add tts_demo.ipynb

examples/audio_utils.py

@@ -0,0 +1,284 @@
+import os
+import io
+import numpy as np
+import pyaudio
+import wave
+import base64
+"""
+    audio utils for modified_funasr_demo.py
+"""
+
+def decode_str2bytes(data):
+    # 将Base64编码的字节串解码为字节串
+    if data is None:
+        return None
+    return base64.b64decode(data.encode('utf-8'))
+
+class BaseAudio:
+    def __init__(self, 
+                 filename=None, 
+                 input=False, 
+                 output=False, 
+                 CHUNK=1024, 
+                 FORMAT=pyaudio.paInt16, 
+                 CHANNELS=1, 
+                 RATE=16000,
+                 input_device_index=None,
+                 output_device_index=None,
+                 **kwargs):
+        self.CHUNK = CHUNK
+        self.FORMAT = FORMAT
+        self.CHANNELS = CHANNELS
+        self.RATE = RATE
+        self.filename = filename
+        assert input!= output, "input and output cannot be the same, \
+            but got input={} and output={}.".format(input, output)
+        print("------------------------------------------")
+        print(f"{'Input' if input else 'Output'} Audio Initialization: ")
+        print(f"CHUNK: {self.CHUNK} \nFORMAT: {self.FORMAT} \nCHANNELS: {self.CHANNELS} \nRATE: {self.RATE} \ninput_device_index: {input_device_index} \noutput_device_index: {output_device_index}")
+        print("------------------------------------------")
+        self.p = pyaudio.PyAudio()
+        self.stream = self.p.open(format=FORMAT,
+                                  channels=CHANNELS,
+                                  rate=RATE,
+                                  input=input,
+                                  output=output,
+                                  input_device_index=input_device_index,
+                                  output_device_index=output_device_index,
+                                  **kwargs)
+    
+    def load_audio_file(self, wav_file):
+        with wave.open(wav_file, 'rb') as wf:
+            params = wf.getparams()
+            frames = wf.readframes(params.nframes)
+            print("Audio file loaded.")
+            # Audio Parameters
+            # print("Channels:", params.nchannels)
+            # print("Sample width:", params.sampwidth)
+            # print("Frame rate:", params.framerate)
+            # print("Number of frames:", params.nframes)
+            # print("Compression type:", params.comptype)
+        return frames
+    
+    def check_audio_type(self, audio_data, return_type=None):
+        assert return_type in ['bytes', 'io', None], \
+            "return_type should be 'bytes', 'io' or None."
+        if isinstance(audio_data, str):
+            if len(audio_data) > 50:
+                audio_data = decode_str2bytes(audio_data)
+            else:
+                assert os.path.isfile(audio_data), \
+                    "audio_data should be a file path or a bytes object."
+                wf = wave.open(audio_data, 'rb')
+                audio_data = wf.readframes(wf.getnframes())
+        elif isinstance(audio_data, np.ndarray):
+            if audio_data.dtype == np.dtype('float32'):
+                audio_data = np.int16(audio_data * np.iinfo(np.int16).max)
+            audio_data = audio_data.tobytes()
+        elif isinstance(audio_data, bytes):
+            pass
+        else:
+            raise TypeError(f"audio_data must be bytes, numpy.ndarray or str, \
+                but got {type(audio_data)}")
+        
+        if return_type == None:
+            return audio_data
+        return self.write_wave(None, [audio_data], return_type)
+    
+    def write_wave(self, filename, frames, return_type='io'):
+        """Write audio data to a file."""
+        if isinstance(frames, bytes):
+            frames = [frames]
+        if not isinstance(frames, list): 
+            raise TypeError("frames should be \
+            a list of bytes or a bytes object, \
+            but got {}.".format(type(frames)))
+        
+        if return_type == 'io':
+            if filename is None:
+                filename = io.BytesIO()
+            if self.filename:
+                filename = self.filename
+            return self.write_wave_io(filename, frames) 
+        elif return_type == 'bytes':
+            return self.write_wave_bytes(frames)
+
+    
+    def write_wave_io(self, filename, frames):
+        """
+        Write audio data to a file-like object.
+        
+        Args:
+            filename: [string or file-like object], file path or file-like object to write
+            frames: list of bytes, audio data to write
+        """
+        wf = wave.open(filename, 'wb')
+        
+        # 设置WAV文件的参数
+        wf.setnchannels(self.CHANNELS)
+        wf.setsampwidth(self.p.get_sample_size(self.FORMAT))
+        wf.setframerate(self.RATE)
+        wf.writeframes(b''.join(frames))
+        wf.close()
+        if isinstance(filename, io.BytesIO):
+            filename.seek(0)   # reset file pointer to beginning
+        return filename
+    
+    def write_wave_bytes(self, frames):
+        """Write audio data to a bytes object."""
+        return b''.join(frames)
+class BaseAudio:
+    def __init__(self, 
+                 filename=None, 
+                 input=False, 
+                 output=False, 
+                 CHUNK=1024, 
+                 FORMAT=pyaudio.paInt16, 
+                 CHANNELS=1, 
+                 RATE=16000,
+                 input_device_index=None,
+                 output_device_index=None,
+                 **kwargs):
+        self.CHUNK = CHUNK
+        self.FORMAT = FORMAT
+        self.CHANNELS = CHANNELS
+        self.RATE = RATE
+        self.filename = filename
+        assert input!= output, "input and output cannot be the same, \
+            but got input={} and output={}.".format(input, output)
+        print("------------------------------------------")
+        print(f"{'Input' if input else 'Output'} Audio Initialization: ")
+        print(f"CHUNK: {self.CHUNK} \nFORMAT: {self.FORMAT} \nCHANNELS: {self.CHANNELS} \nRATE: {self.RATE} \ninput_device_index: {input_device_index} \noutput_device_index: {output_device_index}")
+        print("------------------------------------------")
+        self.p = pyaudio.PyAudio()
+        self.stream = self.p.open(format=FORMAT,
+                                  channels=CHANNELS,
+                                  rate=RATE,
+                                  input=input,
+                                  output=output,
+                                  input_device_index=input_device_index,
+                                  output_device_index=output_device_index,
+                                  **kwargs)
+    
+    def load_audio_file(self, wav_file):
+        with wave.open(wav_file, 'rb') as wf:
+            params = wf.getparams()
+            frames = wf.readframes(params.nframes)
+            print("Audio file loaded.")
+            # Audio Parameters
+            # print("Channels:", params.nchannels)
+            # print("Sample width:", params.sampwidth)
+            # print("Frame rate:", params.framerate)
+            # print("Number of frames:", params.nframes)
+            # print("Compression type:", params.comptype)
+        return frames
+    
+    def check_audio_type(self, audio_data, return_type=None):
+        assert return_type in ['bytes', 'io', None], \
+            "return_type should be 'bytes', 'io' or None."
+        if isinstance(audio_data, str):
+            if len(audio_data) > 50:
+                audio_data = decode_str2bytes(audio_data)
+            else:
+                assert os.path.isfile(audio_data), \
+                    "audio_data should be a file path or a bytes object."
+                wf = wave.open(audio_data, 'rb')
+                audio_data = wf.readframes(wf.getnframes())
+        elif isinstance(audio_data, np.ndarray):
+            if audio_data.dtype == np.dtype('float32'):
+                audio_data = np.int16(audio_data * np.iinfo(np.int16).max)
+            audio_data = audio_data.tobytes()
+        elif isinstance(audio_data, bytes):
+            pass
+        else:
+            raise TypeError(f"audio_data must be bytes, numpy.ndarray or str, \
+                but got {type(audio_data)}")
+        
+        if return_type == None:
+            return audio_data
+        return self.write_wave(None, [audio_data], return_type)
+    
+    def write_wave(self, filename, frames, return_type='io'):
+        """Write audio data to a file."""
+        if isinstance(frames, bytes):
+            frames = [frames]
+        if not isinstance(frames, list): 
+            raise TypeError("frames should be \
+            a list of bytes or a bytes object, \
+            but got {}.".format(type(frames)))
+        
+        if return_type == 'io':
+            if filename is None:
+                filename = io.BytesIO()
+            if self.filename:
+                filename = self.filename
+            return self.write_wave_io(filename, frames) 
+        elif return_type == 'bytes':
+            return self.write_wave_bytes(frames)
+
+    
+    def write_wave_io(self, filename, frames):
+        """
+        Write audio data to a file-like object.
+        
+        Args:
+            filename: [string or file-like object], file path or file-like object to write
+            frames: list of bytes, audio data to write
+        """
+        wf = wave.open(filename, 'wb')
+        
+        # 设置WAV文件的参数
+        wf.setnchannels(self.CHANNELS)
+        wf.setsampwidth(self.p.get_sample_size(self.FORMAT))
+        wf.setframerate(self.RATE)
+        wf.writeframes(b''.join(frames))
+        wf.close()
+        if isinstance(filename, io.BytesIO):
+            filename.seek(0)   # reset file pointer to beginning
+        return filename
+    
+    def write_wave_bytes(self, frames):
+        """Write audio data to a bytes object."""
+        return b''.join(frames)
+
+
+class BaseRecorder(BaseAudio):
+    def __init__(self, 
+                 input=True, 
+                 base_chunk_size=None, 
+                 RATE=16000, 
+                 **kwargs):
+        super().__init__(input=input, RATE=RATE, **kwargs)
+        self.base_chunk_size = base_chunk_size
+        if base_chunk_size is None:
+            self.base_chunk_size = self.CHUNK
+
+    def record(self, 
+               filename,
+               duration=5, 
+               return_type='io',
+               logger=None):
+        if logger is not None:
+            logger.info("Recording started.")
+        else:
+            print("Recording started.")
+        frames = []
+        for i in range(0, int(self.RATE / self.CHUNK * duration)):
+            data = self.stream.read(self.CHUNK, exception_on_overflow=False)
+            frames.append(data)
+        if logger is not None:
+            logger.info("Recording stopped.")
+        else:
+            print("Recording stopped.")
+        return self.write_wave(filename, frames, return_type)
+
+    def record_chunk_voice(self, 
+                           return_type='bytes', 
+                           CHUNK=None, 
+                           exception_on_overflow=True, 
+                           queue=None):
+        data = self.stream.read(self.CHUNK if CHUNK is None else CHUNK, 
+                                exception_on_overflow=exception_on_overflow)
+        if return_type is not None:
+            return self.write_wave(None, [data], return_type)
+        return data

examples/modified_funasr_demo.py

@@ -0,0 +1,39 @@
+import os
+import sys
+sys.path.append(os.path.dirname(os.path.dirname(__file__)))
+
+from audio_utils import BaseRecorder
+from utils.stt.modified_funasr import ModifiedRecognizer
+
+
+
+def asr_file_stream(file_path=r'.\assets\example_recording.wav'):
+    # 读入音频文件
+    rec = BaseRecorder()
+    data = rec.load_audio_file(file_path)
+    
+    # 创建模型
+    asr = ModifiedRecognizer(use_punct=True, use_emotion=True, use_speaker_ver=True)
+    asr.session_signup("test")
+
+    # 记录目标说话人
+    asr.initialize_speaker(r".\assets\example_recording.wav")
+
+    # 语音识别
+    print("===============================================")
+    text_dict = asr.streaming_recognize("test", data, auto_det_end=True)
+    print(f"text_dict: {text_dict}")
+
+    if not isinstance(text_dict, str):
+        print("".join(text_dict['text']))
+    
+    # 情感识别
+    print("===============================================")
+    emotion_dict = asr.recognize_emotion(data)
+    print(f"emotion_dict: {emotion_dict}")
+    if not isinstance(emotion_dict, str):
+        max_index = emotion_dict['scores'].index(max(emotion_dict['scores']))
+        print("emotion: " +emotion_dict['labels'][max_index])
+
+
+asr_file_stream()

examples/speaker_embedding/README.md

@@ -0,0 +1 @@
+存储目标说话人的语音特征,如要修改路径,请修改 utils/stt/speaker_ver_utils中的DEFALUT_SAVE_PATH

examples/tts_demo.ipynb

@@ -0,0 +1,214 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Importing the dtw module. When using in academic works please cite:\n",
+      "  T. Giorgino. Computing and Visualizing Dynamic Time Warping Alignments in R: The dtw Package.\n",
+      "  J. Stat. Soft., doi:10.18637/jss.v031.i07.\n",
+      "\n"
+     ]
+    }
+   ],
+   "source": [
+    "import sys\n",
+    "import os\n",
+    "sys.path.append(\"../\")\n",
+    "from utils.tts.openvoice_utils import TextToSpeech\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\utils\\weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.\n",
+      "  warnings.warn(\"torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.\")\n",
+      "Building prefix dict from the default dictionary ...\n",
+      "Loading model from cache C:\\Users\\bing\\AppData\\Local\\Temp\\jieba.cache\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "load base tts model successfully!\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Loading model cost 0.304 seconds.\n",
+      "Prefix dict has been built successfully.\n",
+      "Some weights of the model checkpoint at bert-base-multilingual-uncased were not used when initializing BertForMaskedLM: ['cls.seq_relationship.bias', 'cls.seq_relationship.weight']\n",
+      "- This IS expected if you are initializing BertForMaskedLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
+      "- This IS NOT expected if you are initializing BertForMaskedLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\modules\\conv.py:797: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
+      "  return F.conv_transpose1d(\n",
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\modules\\conv.py:306: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
+      "  return F.conv1d(input, weight, bias, self.stride,\n",
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\utils\\weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.\n",
+      "  warnings.warn(\"torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.\")\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "generate base speech!\n",
+      "**********************,tts sr 44100\n",
+      "audio segment length is [torch.Size([81565])]\n",
+      "True\n",
+      "Loaded checkpoint 'D:\\python\\OpenVoice\\checkpoints_v2\\converter/checkpoint.pth'\n",
+      "missing/unexpected keys: [] []\n",
+      "load tone color converter successfully!\n"
+     ]
+    }
+   ],
+   "source": [
+    "model = TextToSpeech(use_tone_convert=True, device=\"cuda\", debug=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# 测试用，将mp3转为int32类型的numpy，对齐输入端\n",
+    "from pydub import AudioSegment\n",
+    "import numpy as np\n",
+    "source_audio=r\"D:\\python\\OpenVoice\\resources\\demo_speaker0.mp3\"\n",
+    "audio = AudioSegment.from_file(source_audio, format=\"mp3\")\n",
+    "raw_data = audio.raw_data\n",
+    "audio_array = np.frombuffer(raw_data, dtype=np.int32)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "OpenVoice version: v2\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\functional.py:665: UserWarning: stft with return_complex=False is deprecated. In a future pytorch release, stft will return complex tensors for all inputs, and return_complex=False will raise an error.\n",
+      "Note: you can still call torch.view_as_real on the complex output to recover the old return format. (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\SpectralOps.cpp:878.)\n",
+      "  return _VF.stft(input, n_fft, hop_length, win_length, window,  # type: ignore[attr-defined]\n",
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\modules\\conv.py:456: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
+      "  return F.conv2d(input, weight, bias, self.stride,\n"
+     ]
+    }
+   ],
+   "source": [
+    "# 获取并设置目标说话人的speaker embedding\n",
+    "# audio_array ：输入的音频信号，类型为 np.ndarray\n",
+    "# 获取speaker embedding\n",
+    "target_se = model.audio2emb(audio_array, rate=44100, vad=True)\n",
+    "# 将模型的默认目标说话人embedding设置为 target_se\n",
+    "model.initialize_target_se(target_se)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\modules\\conv.py:797: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
+      "  return F.conv_transpose1d(\n",
+      "c:\\Users\\bing\\.conda\\envs\\openVoice\\lib\\site-packages\\torch\\nn\\modules\\conv.py:306: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
+      "  return F.conv1d(input, weight, bias, self.stride,\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "generate base speech!\n",
+      "**********************,tts sr 44100\n",
+      "audio segment length is [torch.Size([216378])]\n",
+      "Audio saved to D:\\python\\OpenVoice\\outputs_v2\\demo_tts.wav\n"
+     ]
+    }
+   ],
+   "source": [
+    "# 测试base_tts，不含音色转换\n",
+    "text = \"你好呀,我不知道该怎么告诉你这件事，但是我真的很需要你。\"\n",
+    "audio, sr = model._base_tts(text, speed=1)\n",
+    "audio = model.tensor2numpy(audio)\n",
+    "model.save_audio(audio, sr, r\"D:\\python\\OpenVoice\\outputs_v2\\demo_tts.wav\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "generate base speech!\n",
+      "**********************,tts sr 44100\n",
+      "audio segment length is [torch.Size([216378])]\n",
+      "torch.float32\n",
+      "**********************************, convert sr 22050\n",
+      "tone color has been converted!\n",
+      "Audio saved to D:\\python\\OpenVoice\\outputs_v2\\demo.wav\n"
+     ]
+    }
+   ],
+   "source": [
+    "# 测试整体pipeline，包含音色转换\n",
+    "text = \"你好呀,我不知道该怎么告诉你这件事，但是我真的很需要你。\"\n",
+    "audio_bytes, sr = model.tts(text, speed=1)\n",
+    "audio = np.frombuffer(audio_bytes, dtype=np.int16).flatten()\n",
+    "model.save_audio(audio, sr, r\"D:\\python\\OpenVoice\\outputs_v2\\demo.wav\" )"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "openVoice",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.19"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

utils/assets/README.md

@@ -0,0 +1 @@
+./ses 保存 source se 的 embedding 路径，格式为 *.pth 

utils/stt/emotion_utils.py

@@ -0,0 +1,142 @@
+import io
+import numpy as np
+import base64
+import wave
+from funasr import AutoModel
+import time
+"""
+    Base模型
+        不能进行情绪分类,只能用作特征提取
+"""
+FUNASRBASE = {
+    "model_type": "funasr",
+    "model_path": "iic/emotion2vec_base",
+    "model_revision": "v2.0.4"
+}
+
+
+"""
+    Finetune模型
+        输出分类结果
+"""
+FUNASRFINETUNE = {
+    "model_type": "funasr",
+    "model_path": "iic/emotion2vec_base_finetuned"
+}
+
+def decode_str2bytes(data):
+    # 将Base64编码的字节串解码为字节串
+    if data is None:
+        return None
+    return base64.b64decode(data.encode('utf-8'))
+
+class Emotion:
+    def __init__(self,
+                 model_type="funasr",
+                 model_path="iic/emotion2vec_base",
+                 device="cuda",
+                 model_revision="v2.0.4",
+                 **kwargs):
+        
+        self.model_type = model_type
+        self.initialize(model_type, model_path, device, model_revision, **kwargs)
+    
+    # 初始化模型
+    def initialize(self,
+                   model_type,
+                   model_path,
+                   device,
+                   model_revision,
+                   **kwargs):
+        if model_type == "funasr":
+            self.emotion_model = AutoModel(model=model_path, device=device, model_revision=model_revision, **kwargs)
+        else:
+            raise NotImplementedError(f"unsupported model type [{model_type}]. only [funasr] expected.")
+
+    # 检查输入类型
+    def check_audio_type(self,
+                         audio_data):
+        """check audio data type and convert it to bytes if necessary."""
+        if isinstance(audio_data, bytes):
+            pass
+        elif isinstance(audio_data, list):
+            audio_data = b''.join(audio_data)
+        elif isinstance(audio_data, str):
+            audio_data = decode_str2bytes(audio_data)
+        elif isinstance(audio_data, io.BytesIO):
+            wf = wave.open(audio_data, 'rb')
+            audio_data = wf.readframes(wf.getnframes())
+        elif isinstance(audio_data, np.ndarray):
+            pass
+        else:
+            raise TypeError(f"audio_data must be bytes, list, str, \
+                io.BytesIO or numpy array, but got {type(audio_data)}")
+        
+        if isinstance(audio_data, bytes):
+            audio_data = np.frombuffer(audio_data, dtype=np.int16)
+        elif isinstance(audio_data, np.ndarray):
+            if audio_data.dtype != np.int16:
+                audio_data = audio_data.astype(np.int16)
+        else:
+            raise TypeError(f"audio_data must be bytes or numpy array, but got {type(audio_data)}")
+        
+        # 输入类型必须是float32
+        if isinstance(audio_data, np.ndarray):
+            audio_data = audio_data.astype(np.float32)
+        else:
+            raise TypeError(f"audio_data must be numpy array, but got {type(audio_data)}")
+        return audio_data
+    
+    def process(self,
+                audio_data,
+                granularity="utterance",
+                extract_embedding=False,
+                output_dir=None,
+                only_score=True):
+        """
+            audio_data: only float32 expected beacause layernorm
+            extract_embedding: save embedding if true
+            output_dir: save path for embedding
+            only_Score: only return lables & scores if true
+        """
+        audio_data = self.check_audio_type(audio_data)
+        if self.model_type == 'funasr':
+            result = self.emotion_model.generate(audio_data, output_dir=output_dir, granularity=granularity, extract_embedding=extract_embedding)
+        else:
+            pass
+
+        # 只保留 lables 和 scores
+        if only_score:
+            maintain_key = ["labels", "scores"]
+            for res in result:
+                keys_to_remove = [k for k in res.keys() if k not in maintain_key]
+                for k in keys_to_remove:
+                    res.pop(k)
+        return result[0]
+
+# only for test
+def load_audio_file(wav_file):
+    with wave.open(wav_file, 'rb') as wf:
+        params = wf.getparams()
+        frames = wf.readframes(params.nframes)
+        print("Audio file loaded.")
+        # Audio Parameters
+        # print("Channels:", params.nchannels)
+        # print("Sample width:", params.sampwidth)
+        # print("Frame rate:", params.framerate)
+        # print("Number of frames:", params.nframes)
+        # print("Compression type:", params.comptype)
+    return frames
+
+if __name__ == "__main__":
+    inputs = r".\example\test.wav"
+    inputs = load_audio_file(inputs)
+    device = "cuda"
+    # FUNASRBASE.update({"device": device})
+    FUNASRFINETUNE.update({"deivce": device})
+    emotion_model = Emotion(**FUNASRFINETUNE)
+    s = time.time()
+    result = emotion_model.process(inputs)
+    t = time.time()
+    print(t - s)
+    print(result)

utils/stt/modified_funasr.py

@@ -0,0 +1,209 @@
+from .funasr_utils import FunAutoSpeechRecognizer
+from .punctuation_utils import CTTRANSFORMER, Punctuation
+from .emotion_utils import FUNASRFINETUNE, Emotion
+from .speaker_ver_utils import ERES2NETV2, DEFALUT_SAVE_PATH, speaker_verfication
+import os
+
+import numpy as np
+class ModifiedRecognizer(FunAutoSpeechRecognizer):
+    def __init__(self, 
+                 use_punct=True, 
+                 use_emotion=False,
+                 use_speaker_ver=True):
+        
+        # 创建基础的 funasr模型，用于语音识别，识别出不带标点的句子
+        super().__init__(
+                 model_path="paraformer-zh-streaming", 
+                 device="cuda", 
+                 RATE=16000, 
+                 cfg_path=None, 
+                 debug=False, 
+                 chunk_ms=480, 
+                 encoder_chunk_look_back=4, 
+                 decoder_chunk_look_back=1)
+        
+        # 记录是否具备附加功能
+        self.use_punct = use_punct
+        self.use_emotion = use_emotion
+        self.use_speaker_ver = use_speaker_ver
+
+        # 增加标点模型
+        if use_punct:
+            self.puctuation_model = Punctuation(**CTTRANSFORMER)
+        
+        # 情绪识别模型
+        if use_emotion:
+            self.emotion_model = Emotion(**FUNASRFINETUNE)
+
+        # 说话人识别模型
+        if use_speaker_ver:
+            self.speaker_ver_model = speaker_verfication(**ERES2NETV2)
+
+    def initialize_speaker(self, speaker_1_wav):
+        """
+        用于说话人识别，将输入的音频(speaker_1_wav)设立为目标说话人，并将其特征保存本地
+        """
+        if not self.use_speaker_ver:
+            raise NotImplementedError("no access")
+        if speaker_1_wav.endswith(".npy"):
+            self.save_speaker_path = speaker_1_wav
+        elif speaker_1_wav.endswith('.wav'):
+            self.save_speaker_path = os.path.join(DEFALUT_SAVE_PATH, 
+                                          os.path.basename(speaker_1_wav).replace(".wav", ".npy"))
+            # self.save_speaker_path = DEFALUT_SAVE_PATH
+            self.speaker_ver_model.wav2embeddings(speaker_1_wav, self.save_speaker_path)
+        else:
+            raise TypeError("only support [.npy] or [.wav].")
+            
+
+    def speaker_ver(self, speaker_2_wav):
+        """
+        用于说话人识别，判断输入音频是否为目标说话人，
+        是返回True，不是返回False
+        """
+        if not self.use_speaker_ver:
+            raise NotImplementedError("no access")
+        if not hasattr(self, "save_speaker_path"):
+            raise NotImplementedError("please initialize speaker first")
+
+        # self.speaker_ver_model.verfication 返回值为字符串 'yes' / 'no'
+        return self.speaker_ver_model.verfication(base_emb=self.save_speaker_path,
+                                           speaker_2_wav=speaker_2_wav) == 'yes'
+        
+
+    def recognize(self, audio_data):
+        """
+        非流式语音识别，返回识别出的文本，返回值类型 str
+        """
+        audio_data = self.check_audio_type(audio_data)
+
+        # 说话人识别
+        if self.use_speaker_ver:
+            if self.speaker_ver_model.verfication(self.save_speaker_path, 
+                                                  speaker_2_wav=audio_data) == 'no':
+                return "Other People"
+
+        # 语音识别
+        result = self.asr_model.generate(input=audio_data,
+                                         batch_size_s=300,
+                                         hotword=self.hotwords)
+        text = ''
+        for res in result:
+            text += res['text']
+
+        # 添加标点
+        if self.use_punct:
+            text = self.puctuation_model.process(text+'#', append_period=False).replace('#', '')
+
+        return text
+    
+    def recognize_emotion(self, audio_data):
+        """
+        情感识别，返回值为:
+            1. 如果说话人非目标说话人，返回字符串 "Other People"
+            2. 如果说话人为目标说话人，返回字典{"Labels": List[str], "scores": List[int]}
+        """
+        audio_data = self.check_audio_type(audio_data)
+
+        if self.use_speaker_ver:
+            if self.speaker_ver_model.verfication(self.save_speaker_path, 
+                                                  speaker_2_wav=audio_data) == 'no':
+                return "Other People"
+            
+        if self.use_emotion:
+            return self.emotion_model.process(audio_data)
+        else:
+            raise NotImplementedError("no access")
+
+    def streaming_recognize(self, session_id, audio_data, is_end=False, auto_det_end=False):
+        """recognize partial result
+        
+        Args:
+            audio_data: bytes or numpy array, partial audio data
+            is_end: bool, whether the audio data is the end of a sentence
+            auto_det_end: bool, whether to automatically detect the end of a audio data
+
+        流式语音识别，返回值为：
+            1. 如果说话人非目标说话人，返回字符串 "Other People"
+            2. 如果说话人为目标说话人，返回字典{"test": List[str], "is_end": boolean}
+        """
+        audio_cache = self.audio_cache[session_id]
+        asr_cache = self.asr_cache[session_id]
+        text_dict = dict(text=[], is_end=is_end)
+        audio_data = self.check_audio_type(audio_data)
+        
+        # 说话人识别
+        if self.use_speaker_ver:
+            if self.speaker_ver_model.verfication(self.save_speaker_path, 
+                                                  speaker_2_wav=audio_data) == 'no':
+                return "Other People"
+        
+        # 语音识别
+        if audio_cache is None:
+            audio_cache = audio_data
+        else:
+            # print(f"audio_data: {audio_data.shape}, audio_cache: {self.audio_cache.shape}")
+            if audio_cache.shape[0] > 0:
+                audio_cache = np.concatenate([audio_cache, audio_data], axis=0)
+        
+        if not is_end and audio_cache.shape[0] < self.chunk_partial_size:
+            self.audio_cache[session_id] = audio_cache
+            return text_dict
+        
+        total_chunk_num = int((len(self.audio_cache)-1)/self.chunk_partial_size)
+        
+        if is_end:
+            # if the audio data is the end of a sentence, \
+            # we need to add one more chunk to the end to \
+            # ensure the end of the sentence is recognized correctly.
+            auto_det_end = True
+        
+        if auto_det_end:
+            total_chunk_num += 1
+
+        # print(f"chunk_size: {self.chunk_size}, chunk_stride: {self.chunk_partial_size}, total_chunk_num: {total_chunk_num}, len: {len(self.audio_cache)}")
+        end_idx = None
+        for i in range(total_chunk_num):
+            if auto_det_end:
+                is_end = i == total_chunk_num - 1
+            start_idx = i*self.chunk_partial_size
+            if auto_det_end:
+                end_idx = (i+1)*self.chunk_partial_size if i < total_chunk_num-1 else -1
+            else:
+                end_idx = (i+1)*self.chunk_partial_size if i < total_chunk_num else -1
+            # print(f"cut part: {start_idx}:{end_idx}, is_end: {is_end}, i: {i}, total_chunk_num: {total_chunk_num}")
+            # t_stamp = time.time()
+            
+            speech_chunk = audio_cache[start_idx:end_idx]
+
+            # TODO: exceptions processes
+            try:
+                res = self.asr_model.generate(input=speech_chunk, cache=asr_cache, is_final=is_end, chunk_size=self.chunk_size, encoder_chunk_look_back=self.encoder_chunk_look_back, decoder_chunk_look_back=self.decoder_chunk_look_back)
+            except ValueError as e:
+                print(f"ValueError: {e}")
+                continue
+            
+            # 增添标点
+            if self.use_punct:
+                text_dict['text'].append(self.puctuation_model.process(self.text_postprecess(res[0], data_id='text'), cache=text_dict))
+            else:
+                text_dict['text'].append(self.text_postprecess(res[0], data_id='text'))
+
+
+            # print(f"each chunk time: {time.time()-t_stamp}")
+            
+        if is_end:
+            audio_cache = None
+            asr_cache = {}
+        else:
+            if end_idx:
+                audio_cache = self.audio_cache[end_idx:] # cut the processed part from audio_cache
+        text_dict['is_end'] = is_end
+
+        if self.use_punct and is_end:
+            text_dict['text'].append(self.puctuation_model.process('#', cache=text_dict).replace('#', ''))
+            
+        self.audio_cache[session_id] = audio_cache
+        self.asr_cache[session_id] = asr_cache
+        # print(f"text_dict: {text_dict}")
+        return text_dict

utils/stt/punctuation_utils.py

@@ -0,0 +1,119 @@
+from funasr import AutoModel
+from modelscope.pipelines import pipeline
+from modelscope.utils.constant import Tasks
+
+PUNCTUATION_MARK = [",", ".", "?", "!", "，", "。", "？", "！"]
+"""
+FUNASR
+    模型大小: 1G
+    效果: 较好
+    输入类型: 仅支持字符串不支持list, 输入list会将list视为彼此独立的字符串处理
+"""
+FUNASR = {
+    "model_type": "funasr", 
+    "model_path": "ct-punc",
+    "model_revision": "v2.0.4"
+}
+"""
+CTTRANSFORMER
+    模型大小: 275M
+    效果：较差
+    输入类型: 支持字符串与list, 同时支持输入cache
+"""
+CTTRANSFORMER = {
+    "model_type": "ct-transformer",
+    "model_path": "iic/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
+    "model_revision": "v2.0.4"
+}
+
+class Punctuation:
+    def __init__(self,
+                 model_type="funasr", # funasr | ct-transformer
+                 model_path="ct-punc",
+                 device="cuda",
+                 model_revision="v2.0.4",
+                 **kwargs):
+        
+        self.model_type=model_type
+        self.initialize(model_type, model_path, device, model_revision, **kwargs)
+
+    def initialize(self,
+                   model_type,
+                   model_path,
+                   device,
+                   model_revision,
+                   **kwargs):
+        if model_type == 'funasr':
+            self.punc_model = AutoModel(model=model_path, device=device, model_revision=model_revision, **kwargs)
+        elif model_type == 'ct-transformer':
+            self.punc_model = pipeline(task=Tasks.punctuation, model=model_path, model_revision=model_revision, **kwargs)
+        else:
+            raise NotImplementedError(f"unsupported model type [{model_type}]. only [funasr|ct-transformer] expected.")
+
+    def check_text_type(self,
+                        text_data):
+        # funasr只支持单个str输入，不支持list输入，此处将list转化为字符串
+        if self.model_type == 'funasr':
+            if isinstance(text_data, str):
+                pass
+            elif isinstance(text_data, list):
+                text_data = ''.join(text_data)
+            else:
+                raise TypeError(f"text must be str or list, but got {type(list)}")
+        # ct-transformer支持list输入
+        # TODO 验证拆分字符串能否提高效率
+        elif self.model_type == 'ct-transformer':
+            if isinstance(text_data, str):
+                text_data = [text_data]
+            elif isinstance(text_data, list):
+                pass
+            else:
+                raise TypeError(f"text must be str or list, but got {type(list)}")
+        else:
+            pass
+        return text_data
+
+    def generate_cache(self, cache):
+        new_cache = {'pre_text': ""}
+        for text in cache['text']:
+            if text != '':
+                new_cache['pre_text'] = new_cache['pre_text']+text
+        return new_cache
+
+    def process(self,
+                text,
+                append_period=False,
+                cache={}):
+        if text == '':
+            return ''
+        text = self.check_text_type(text)
+        if self.model_type == 'funasr':
+            result = self.punc_model.generate(text)
+        elif self.model_type == 'ct-transformer':
+            if cache != {}:
+                cache = self.generate_cache(cache)
+            result = self.punc_model(text, cache=cache)
+        punced_text = ''
+        for res in result:
+            punced_text += res['text']
+        # 如果最后没有标点符号，手动加上。
+        if append_period and not punced_text[-1] in PUNCTUATION_MARK:
+            punced_text += "。"
+        return punced_text
+
+if __name__ == "__main__":
+    inputs = "把字符串拆分为list只|适用于ct-transformer模型|在数据处理部分|已经把list转为单个字符串"
+    """
+    把字符串拆分为list只适用于ct-transformer模型,
+    在数据处理部分,已经把list转为单个字符串
+    """
+    vads = inputs.split("|")
+    device = "cuda"
+    CTTRANSFORMER.update({"device": device})
+    puct_model = Punctuation(**CTTRANSFORMER)
+    result = puct_model.process(vads)
+    print(result)
+    # FUNASR.update({"device":"cuda"})
+    # puct_model = Punctuation(**FUNASR)   
+    # result = puct_model.process(vads)
+    # print(result)

utils/stt/speaker_ver_utils.py

@@ -0,0 +1,86 @@
+from modelscope.pipelines import pipeline
+import numpy as np
+import os
+
+ERES2NETV2 = {
+    "task": 'speaker-verification',
+    "model_name": 'damo/speech_eres2netv2_sv_zh-cn_16k-common',
+    "model_revision": 'v1.0.1',
+    "save_embeddings": False
+}
+
+# 保存 embedding 的路径
+DEFALUT_SAVE_PATH = os.path.join(os.path.dirname(os.path.dirname(__name__)), "speaker_embedding")
+
+class speaker_verfication:
+    def __init__(self, 
+                task='speaker-verification',
+                model_name='damo/speech_eres2netv2_sv_zh-cn_16k-common',
+                model_revision='v1.0.1',
+                device="cuda",
+                save_embeddings=False):
+        self.pipeline = pipeline(
+                task=task,
+                model=model_name,
+                model_revision=model_revision,
+                device=device)
+        self.save_embeddings = save_embeddings
+
+    def wav2embeddings(self, speaker_1_wav, save_path=None):
+        result = self.pipeline([speaker_1_wav], output_emb=True)
+        speaker_1_emb = result['embs'][0]
+        if save_path is not None:
+            np.save(save_path, speaker_1_emb)
+        return speaker_1_emb
+
+    def _verifaction(self, speaker_1_wav, speaker_2_wav, threshold, save_path):
+        if not self.save_embeddings:
+            result = self.pipeline([speaker_1_wav, speaker_2_wav], thr=threshold)
+            return result["text"]
+        else:
+            result = self.pipeline([speaker_1_wav, speaker_2_wav], thr=threshold, output_emb=True)
+            speaker1_emb = result["embs"][0]
+            speaker2_emb = result["embs"][1]
+            np.save(os.path.join(save_path, "speaker_1.npy"), speaker1_emb)
+            return result['outputs']["text"]
+        
+    def _verifaction_from_embedding(self, base_emb, speaker_2_wav, threshold):
+        base_emb = np.load(base_emb)
+        result = self.pipeline([speaker_2_wav], output_emb=True)
+        speaker2_emb = result["embs"][0]
+        similarity = np.dot(base_emb, speaker2_emb) / (np.linalg.norm(base_emb) * np.linalg.norm(speaker2_emb))
+        if similarity > threshold:
+            return "yes"
+        else:
+            return "no"
+    
+    def verfication(self, 
+                    base_emb=None, 
+                    speaker_1_wav=None, 
+                    speaker_2_wav=None, 
+                    threshold=0.333, 
+                    save_path=None):
+        if base_emb is not None and speaker_1_wav is not None:
+            raise ValueError("Only need one of them, base_emb or speaker_1_wav")
+        if base_emb is not None and speaker_2_wav is not None:
+            return self._verifaction_from_embedding(base_emb, speaker_2_wav, threshold)
+        elif speaker_1_wav is not None and speaker_2_wav is not None:
+            return self._verifaction(speaker_1_wav, speaker_2_wav, threshold, save_path)
+        else:
+            raise NotImplementedError
+        
+if __name__ == '__main__':
+    verifier = speaker_verfication(**ERES2NETV2)
+
+    verifier = speaker_verfication(save_embeddings=False)
+    result = verifier.verfication(base_emb=None, speaker_1_wav=r"C:\Users\bing\Downloads\speaker1_a_cn_16k.wav", 
+                                  speaker_2_wav=r"C:\Users\bing\Downloads\speaker2_a_cn_16k.wav",
+                                  threshold=0.333,
+                                  save_path=r"D:\python\irving\takway_base-main\savePath"
+                                  )
+    print("---")
+    print(result)
+    print(verifier.verfication(r"D:\python\irving\takway_base-main\savePath\speaker_1.npy",
+                               speaker_2_wav=r"C:\Users\bing\Downloads\speaker1_b_cn_16k.wav",
+                               threshold=0.333,
+                               ))

utils/tts/TTS_README.md

@@ -0,0 +1,7 @@
+1. 安装 melo
+   参考 https://github.com/myshell-ai/OpenVoice/blob/main/docs/USAGE.md#openvoice-v2
+2. 修改 openvoice_utils.py 中的路径 (包括 SOURCE_SE_DIR / CACHE_PATH / OPENVOICE_TONE_COLOR_CONVERTER.converter_path )
+   其中:
+    SOURCE_SE_DIR 改为 utils/assets/ses
+    OPENVOICE_TONE_COLOR_CONVERTER.converter_path 修改为下载到的模型参数路径，下载链接为 https://myshell-public-repo-hosting.s3.amazonaws.com/openvoice/checkpoints_v2_0417.zip
+3. 参考 examples/tts_demo.ipynb 进行代码迁移

utils/tts/openvoice/api.py

@@ -0,0 +1,202 @@
+import torch
+import numpy as np
+import re
+import soundfile
+from utils.tts.openvoice import utils
+from utils.tts.openvoice import commons
+import os
+import librosa
+from utils.tts.openvoice.text import text_to_sequence
+from utils.tts.openvoice.mel_processing import spectrogram_torch
+from utils.tts.openvoice.models import SynthesizerTrn
+
+
+class OpenVoiceBaseClass(object):
+    def __init__(self, 
+                config_path, 
+                device='cuda:0'):
+        if 'cuda' in device:
+            assert torch.cuda.is_available()
+
+        hps = utils.get_hparams_from_file(config_path)
+
+        model = SynthesizerTrn(
+            len(getattr(hps, 'symbols', [])),
+            hps.data.filter_length // 2 + 1,
+            n_speakers=hps.data.n_speakers,
+            **hps.model,
+        ).to(device)
+
+        model.eval()
+        self.model = model
+        self.hps = hps
+        self.device = device
+
+    def load_ckpt(self, ckpt_path):
+        checkpoint_dict = torch.load(ckpt_path, map_location=torch.device(self.device))
+        a, b = self.model.load_state_dict(checkpoint_dict['model'], strict=False)
+        print("Loaded checkpoint '{}'".format(ckpt_path))
+        print('missing/unexpected keys:', a, b)
+
+
+class BaseSpeakerTTS(OpenVoiceBaseClass):
+    language_marks = {
+        "english": "EN",
+        "chinese": "ZH",
+    }
+
+    @staticmethod
+    def get_text(text, hps, is_symbol):
+        text_norm = text_to_sequence(text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
+        if hps.data.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    @staticmethod
+    def audio_numpy_concat(segment_data_list, sr, speed=1.):
+        audio_segments = []
+        for segment_data in segment_data_list:
+            audio_segments += segment_data.reshape(-1).tolist()
+            audio_segments += [0] * int((sr * 0.05)/speed)
+        audio_segments = np.array(audio_segments).astype(np.float32)
+        return audio_segments
+
+    @staticmethod
+    def split_sentences_into_pieces(text, language_str):
+        texts = utils.split_sentence(text, language_str=language_str)
+        print(" > Text splitted to sentences.")
+        print('\n'.join(texts))
+        print(" > ===========================")
+        return texts
+
+    def tts(self, text, output_path, speaker, language='English', speed=1.0):
+        mark = self.language_marks.get(language.lower(), None)
+        assert mark is not None, f"language {language} is not supported"
+
+        texts = self.split_sentences_into_pieces(text, mark)
+
+        audio_list = []
+        for t in texts:
+            t = re.sub(r'([a-z])([A-Z])', r'\1 \2', t)
+            t = f'[{mark}]{t}[{mark}]'
+            stn_tst = self.get_text(t, self.hps, False)
+            device = self.device
+            speaker_id = self.hps.speakers[speaker]
+            with torch.no_grad():
+                x_tst = stn_tst.unsqueeze(0).to(device)
+                x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(device)
+                sid = torch.LongTensor([speaker_id]).to(device)
+                audio = self.model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=0.667, noise_scale_w=0.6,
+                                    length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
+            audio_list.append(audio)
+        audio = self.audio_numpy_concat(audio_list, sr=self.hps.data.sampling_rate, speed=speed)
+
+        if output_path is None:
+            return audio
+        else:
+            soundfile.write(output_path, audio, self.hps.data.sampling_rate)
+
+
+class ToneColorConverter(OpenVoiceBaseClass):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        if kwargs.get('enable_watermark', True):
+            import wavmark
+            self.watermark_model = wavmark.load_model().to(self.device)
+        else:
+            self.watermark_model = None
+        self.version = getattr(self.hps, '_version_', "v1")
+
+
+
+    def extract_se(self, ref_wav_list, se_save_path=None):
+        if isinstance(ref_wav_list, str):
+            ref_wav_list = [ref_wav_list]
+        
+        device = self.device
+        hps = self.hps
+        gs = []
+        
+        for fname in ref_wav_list:
+            audio_ref, sr = librosa.load(fname, sr=hps.data.sampling_rate)
+            y = torch.FloatTensor(audio_ref)
+            y = y.to(device)
+            y = y.unsqueeze(0)
+            y = spectrogram_torch(y, hps.data.filter_length,
+                                        hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length,
+                                        center=False).to(device)
+            with torch.no_grad():
+                g = self.model.ref_enc(y.transpose(1, 2)).unsqueeze(-1)
+                gs.append(g.detach())
+        gs = torch.stack(gs).mean(0)
+
+        if se_save_path is not None:
+            os.makedirs(os.path.dirname(se_save_path), exist_ok=True)
+            torch.save(gs.cpu(), se_save_path)
+
+        return gs
+
+    def convert(self, audio_src_path, src_se, tgt_se, output_path=None, tau=0.3, message="default"):
+        hps = self.hps
+        # load audio
+        audio, sample_rate = librosa.load(audio_src_path, sr=hps.data.sampling_rate)
+        audio = torch.tensor(audio).float()
+        
+        with torch.no_grad():
+            y = torch.FloatTensor(audio).to(self.device)
+            y = y.unsqueeze(0)
+            spec = spectrogram_torch(y, hps.data.filter_length,
+                                    hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length,
+                                    center=False).to(self.device)
+            spec_lengths = torch.LongTensor([spec.size(-1)]).to(self.device)
+            audio = self.model.voice_conversion(spec, spec_lengths, sid_src=src_se, sid_tgt=tgt_se, tau=tau)[0][
+                        0, 0].data.cpu().float().numpy()
+            audio = self.add_watermark(audio, message)
+            if output_path is None:
+                return audio
+            else:
+                soundfile.write(output_path, audio, hps.data.sampling_rate)
+    
+    def add_watermark(self, audio, message):
+        if self.watermark_model is None:
+            return audio
+        device = self.device
+        bits = utils.string_to_bits(message).reshape(-1)
+        n_repeat = len(bits) // 32
+
+        K = 16000
+        coeff = 2
+        for n in range(n_repeat):
+            trunck = audio[(coeff * n) * K: (coeff * n + 1) * K]
+            if len(trunck) != K:
+                print('Audio too short, fail to add watermark')
+                break
+            message_npy = bits[n * 32: (n + 1) * 32]
+            
+            with torch.no_grad():
+                signal = torch.FloatTensor(trunck).to(device)[None]
+                message_tensor = torch.FloatTensor(message_npy).to(device)[None]
+                signal_wmd_tensor = self.watermark_model.encode(signal, message_tensor)
+                signal_wmd_npy = signal_wmd_tensor.detach().cpu().squeeze()
+            audio[(coeff * n) * K: (coeff * n + 1) * K] = signal_wmd_npy
+        return audio
+
+    def detect_watermark(self, audio, n_repeat):
+        bits = []
+        K = 16000
+        coeff = 2
+        for n in range(n_repeat):
+            trunck = audio[(coeff * n) * K: (coeff * n + 1) * K]
+            if len(trunck) != K:
+                print('Audio too short, fail to detect watermark')
+                return 'Fail'
+            with torch.no_grad():
+                signal = torch.FloatTensor(trunck).to(self.device).unsqueeze(0)
+                message_decoded_npy = (self.watermark_model.decode(signal) >= 0.5).int().detach().cpu().numpy().squeeze()
+            bits.append(message_decoded_npy)
+        bits = np.stack(bits).reshape(-1, 8)
+        message = utils.bits_to_string(bits)
+        return message
+    

utils/tts/openvoice/attentions.py

@@ -0,0 +1,465 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from utils.tts.openvoice import commons
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=4,
+        isflow=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        # if isflow:
+        #  cond_layer = torch.nn.Conv1d(256, 2*hidden_channels*n_layers, 1)
+        #  self.cond_pre = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, 1)
+        #  self.cond_layer = weight_norm(cond_layer, name='weight')
+        #  self.gin_channels = 256
+        self.cond_layer_idx = self.n_layers
+        if "gin_channels" in kwargs:
+            self.gin_channels = kwargs["gin_channels"]
+            if self.gin_channels != 0:
+                self.spk_emb_linear = nn.Linear(self.gin_channels, self.hidden_channels)
+                # vits2 says 3rd block, so idx is 2 by default
+                self.cond_layer_idx = (
+                    kwargs["cond_layer_idx"] if "cond_layer_idx" in kwargs else 2
+                )
+                # logging.debug(self.gin_channels, self.cond_layer_idx)
+                assert (
+                    self.cond_layer_idx < self.n_layers
+                ), "cond_layer_idx should be less than n_layers"
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, g=None):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            if i == self.cond_layer_idx and g is not None:
+                g = self.spk_emb_linear(g.transpose(1, 2))
+                g = g.transpose(1, 2)
+                x = x + g
+                x = x * x_mask
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # pad along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

utils/tts/openvoice/commons.py

@@ -0,0 +1,160 @@
+import math
+import torch
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def intersperse(lst, item):
+    result = [item] * (len(lst) * 2 + 1)
+    result[1::2] = lst
+    return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

utils/tts/openvoice/mel_processing.py

@@ -0,0 +1,183 @@
+import torch
+import torch.utils.data
+from librosa.filters import mel as librosa_mel_fn
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    output = dynamic_range_compression_torch(magnitudes)
+    return output
+
+
+def spectral_de_normalize_torch(magnitudes):
+    output = dynamic_range_decompression_torch(magnitudes)
+    return output
+
+
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    if torch.min(y) < -1.1:
+        print("min value is ", torch.min(y))
+    if torch.max(y) > 1.1:
+        print("max value is ", torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + "_" + str(y.device)
+    wnsize_dtype_device = str(win_size) + "_" + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device
+        )
+
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1),
+        (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+        mode="reflect",
+    )
+    y = y.squeeze(1)
+
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spectrogram_torch_conv(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    # if torch.min(y) < -1.:
+    #     print('min value is ', torch.min(y))
+    # if torch.max(y) > 1.:
+    #     print('max value is ', torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    
+    # ******************** original ************************#
+    # y = y.squeeze(1)
+    # spec1 = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+    #                   center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+
+    # ******************** ConvSTFT ************************#
+    freq_cutoff = n_fft // 2 + 1
+    fourier_basis = torch.view_as_real(torch.fft.fft(torch.eye(n_fft)))
+    forward_basis = fourier_basis[:freq_cutoff].permute(2, 0, 1).reshape(-1, 1, fourier_basis.shape[1])
+    forward_basis = forward_basis * torch.as_tensor(librosa.util.pad_center(torch.hann_window(win_size), size=n_fft)).float()
+
+    import torch.nn.functional as F
+
+    # if center:
+    #     signal = F.pad(y[:, None, None, :], (n_fft // 2, n_fft // 2, 0, 0), mode = 'reflect').squeeze(1)
+    assert center is False
+
+    forward_transform_squared = F.conv1d(y, forward_basis.to(y.device), stride = hop_size)
+    spec2 = torch.stack([forward_transform_squared[:, :freq_cutoff, :], forward_transform_squared[:, freq_cutoff:, :]], dim = -1)
+
+
+    # ******************** Verification ************************#
+    spec1 = torch.stft(y.squeeze(1), n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+    assert torch.allclose(spec1, spec2, atol=1e-4)
+
+    spec = torch.sqrt(spec2.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    global mel_basis
+    dtype_device = str(spec.dtype) + "_" + str(spec.device)
+    fmax_dtype_device = str(fmax) + "_" + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=spec.dtype, device=spec.device
+        )
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+    return spec
+
+
+def mel_spectrogram_torch(
+    y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False
+):
+    if torch.min(y) < -1.0:
+        print("min value is ", torch.min(y))
+    if torch.max(y) > 1.0:
+        print("max value is ", torch.max(y))
+
+    global mel_basis, hann_window
+    dtype_device = str(y.dtype) + "_" + str(y.device)
+    fmax_dtype_device = str(fmax) + "_" + dtype_device
+    wnsize_dtype_device = str(win_size) + "_" + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=y.dtype, device=y.device
+        )
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device
+        )
+
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1),
+        (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+        mode="reflect",
+    )
+    y = y.squeeze(1)
+
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+
+    return spec

utils/tts/openvoice/models.py

@@ -0,0 +1,499 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from utils.tts.openvoice import commons
+from utils.tts.openvoice import modules
+from utils.tts.openvoice import attentions
+
+from torch.nn import Conv1d, ConvTranspose1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+
+from utils.tts.openvoice.commons import init_weights, get_padding
+
+
+class TextEncoder(nn.Module):
+	def __init__(self,
+			n_vocab,
+			out_channels,
+			hidden_channels,
+			filter_channels,
+			n_heads,
+			n_layers,
+			kernel_size,
+			p_dropout):
+		super().__init__()
+		self.n_vocab = n_vocab
+		self.out_channels = out_channels
+		self.hidden_channels = hidden_channels
+		self.filter_channels = filter_channels
+		self.n_heads = n_heads
+		self.n_layers = n_layers
+		self.kernel_size = kernel_size
+		self.p_dropout = p_dropout
+
+		self.emb = nn.Embedding(n_vocab, hidden_channels)
+		nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
+
+		self.encoder = attentions.Encoder(
+			hidden_channels,
+			filter_channels,
+			n_heads,
+			n_layers,
+			kernel_size,
+			p_dropout)
+		self.proj= nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+	def forward(self, x, x_lengths):
+		x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h]
+		x = torch.transpose(x, 1, -1) # [b, h, t]
+		x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
+
+		x = self.encoder(x * x_mask, x_mask)
+		stats = self.proj(x) * x_mask
+
+		m, logs = torch.split(stats, self.out_channels, dim=1)
+		return x, m, logs, x_mask
+     
+
+class DurationPredictor(nn.Module):
+    def __init__(
+        self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0
+    ):
+        super().__init__()
+
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.gin_channels = gin_channels
+
+        self.drop = nn.Dropout(p_dropout)
+        self.conv_1 = nn.Conv1d(
+            in_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_1 = modules.LayerNorm(filter_channels)
+        self.conv_2 = nn.Conv1d(
+            filter_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_2 = modules.LayerNorm(filter_channels)
+        self.proj = nn.Conv1d(filter_channels, 1, 1)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, in_channels, 1)
+
+    def forward(self, x, x_mask, g=None):
+        x = torch.detach(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.conv_1(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_1(x)
+        x = self.drop(x)
+        x = self.conv_2(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_2(x)
+        x = self.drop(x)
+        x = self.proj(x * x_mask)
+        return x * x_mask
+               
+class StochasticDurationPredictor(nn.Module):
+	def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, n_flows=4, gin_channels=0):
+		super().__init__()
+		filter_channels = in_channels # it needs to be removed from future version.
+		self.in_channels = in_channels
+		self.filter_channels = filter_channels
+		self.kernel_size = kernel_size
+		self.p_dropout = p_dropout
+		self.n_flows = n_flows
+		self.gin_channels = gin_channels
+
+		self.log_flow = modules.Log()
+		self.flows = nn.ModuleList()
+		self.flows.append(modules.ElementwiseAffine(2))
+		for i in range(n_flows):
+			self.flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+			self.flows.append(modules.Flip())
+
+		self.post_pre = nn.Conv1d(1, filter_channels, 1)
+		self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1)
+		self.post_convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
+		self.post_flows = nn.ModuleList()
+		self.post_flows.append(modules.ElementwiseAffine(2))
+		for i in range(4):
+			self.post_flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+			self.post_flows.append(modules.Flip())
+
+		self.pre = nn.Conv1d(in_channels, filter_channels, 1)
+		self.proj = nn.Conv1d(filter_channels, filter_channels, 1)
+		self.convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
+		if gin_channels != 0:
+			self.cond = nn.Conv1d(gin_channels, filter_channels, 1)
+
+	def forward(self, x, x_mask, w=None, g=None, reverse=False, noise_scale=1.0):
+		x = torch.detach(x)
+		x = self.pre(x)
+		if g is not None:
+			g = torch.detach(g)
+			x = x + self.cond(g)
+		x = self.convs(x, x_mask)
+		x = self.proj(x) * x_mask
+
+		if not reverse:
+			flows = self.flows
+			assert w is not None
+
+			logdet_tot_q = 0
+			h_w = self.post_pre(w)
+			h_w = self.post_convs(h_w, x_mask)
+			h_w = self.post_proj(h_w) * x_mask
+			e_q = torch.randn(w.size(0), 2, w.size(2)).to(device=x.device, dtype=x.dtype) * x_mask
+			z_q = e_q
+			for flow in self.post_flows:
+				z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w))
+				logdet_tot_q += logdet_q
+			z_u, z1 = torch.split(z_q, [1, 1], 1)
+			u = torch.sigmoid(z_u) * x_mask
+			z0 = (w - u) * x_mask
+			logdet_tot_q += torch.sum((F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1,2])
+			logq = torch.sum(-0.5 * (math.log(2*math.pi) + (e_q**2)) * x_mask, [1,2]) - logdet_tot_q
+
+			logdet_tot = 0
+			z0, logdet = self.log_flow(z0, x_mask)
+			logdet_tot += logdet
+			z = torch.cat([z0, z1], 1)
+			for flow in flows:
+				z, logdet = flow(z, x_mask, g=x, reverse=reverse)
+				logdet_tot = logdet_tot + logdet
+			nll = torch.sum(0.5 * (math.log(2*math.pi) + (z**2)) * x_mask, [1,2]) - logdet_tot
+			return nll + logq # [b]
+		else:
+			flows = list(reversed(self.flows))
+			flows = flows[:-2] + [flows[-1]] # remove a useless vflow
+			z = torch.randn(x.size(0), 2, x.size(2)).to(device=x.device, dtype=x.dtype) * noise_scale
+			for flow in flows:
+				z = flow(z, x_mask, g=x, reverse=reverse)
+			z0, z1 = torch.split(z, [1, 1], 1)
+			logw = z0
+			return logw
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None, tau=1.0):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * tau * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        print("Removing weight norm...")
+        for layer in self.ups:
+            remove_weight_norm(layer)
+        for layer in self.resblocks:
+            layer.remove_weight_norm()
+
+
+class ReferenceEncoder(nn.Module):
+    """
+    inputs --- [N, Ty/r, n_mels*r]  mels
+    outputs --- [N, ref_enc_gru_size]
+    """
+
+    def __init__(self, spec_channels, gin_channels=0, layernorm=True):
+        super().__init__()
+        self.spec_channels = spec_channels
+        ref_enc_filters = [32, 32, 64, 64, 128, 128]
+        K = len(ref_enc_filters)
+        filters = [1] + ref_enc_filters
+        convs = [
+            weight_norm(
+                nn.Conv2d(
+                    in_channels=filters[i],
+                    out_channels=filters[i + 1],
+                    kernel_size=(3, 3),
+                    stride=(2, 2),
+                    padding=(1, 1),
+                )
+            )
+            for i in range(K)
+        ]
+        self.convs = nn.ModuleList(convs)
+
+        out_channels = self.calculate_channels(spec_channels, 3, 2, 1, K)
+        self.gru = nn.GRU(
+            input_size=ref_enc_filters[-1] * out_channels,
+            hidden_size=256 // 2,
+            batch_first=True,
+        )
+        self.proj = nn.Linear(128, gin_channels)
+        if layernorm:
+            self.layernorm = nn.LayerNorm(self.spec_channels)
+        else:
+            self.layernorm = None
+
+    def forward(self, inputs, mask=None):
+        N = inputs.size(0)
+
+        out = inputs.view(N, 1, -1, self.spec_channels)  # [N, 1, Ty, n_freqs]
+        if self.layernorm is not None:
+            out = self.layernorm(out)
+
+        for conv in self.convs:
+            out = conv(out)
+            # out = wn(out)
+            out = F.relu(out)  # [N, 128, Ty//2^K, n_mels//2^K]
+
+        out = out.transpose(1, 2)  # [N, Ty//2^K, 128, n_mels//2^K]
+        T = out.size(1)
+        N = out.size(0)
+        out = out.contiguous().view(N, T, -1)  # [N, Ty//2^K, 128*n_mels//2^K]
+
+        self.gru.flatten_parameters()
+        memory, out = self.gru(out)  # out --- [1, N, 128]
+
+        return self.proj(out.squeeze(0))
+
+    def calculate_channels(self, L, kernel_size, stride, pad, n_convs):
+        for i in range(n_convs):
+            L = (L - kernel_size + 2 * pad) // stride + 1
+        return L
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(self,
+            channels,
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            n_flows=4,
+            gin_channels=0):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(modules.ResidualCouplingLayer(channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True))
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+class SynthesizerTrn(nn.Module):
+    """
+    Synthesizer for Training
+    """
+
+    def __init__(
+        self,
+        n_vocab,
+        spec_channels,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        n_speakers=256,
+        gin_channels=256,
+        zero_g=False,
+        **kwargs
+    ):
+        super().__init__()
+
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+
+        self.flow = ResidualCouplingBlock(inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels)
+
+        self.n_speakers = n_speakers
+        if n_speakers == 0:
+            self.ref_enc = ReferenceEncoder(spec_channels, gin_channels)
+        else:
+            self.enc_p = TextEncoder(n_vocab,
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout)
+            self.sdp = StochasticDurationPredictor(hidden_channels, 192, 3, 0.5, 4, gin_channels=gin_channels)
+            self.dp = DurationPredictor(hidden_channels, 256, 3, 0.5, gin_channels=gin_channels)
+            self.emb_g = nn.Embedding(n_speakers, gin_channels)
+        self.zero_g = zero_g
+
+    def infer(self, x, x_lengths, sid=None, noise_scale=1, length_scale=1, noise_scale_w=1., sdp_ratio=0.2, max_len=None):
+        x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths)
+        if self.n_speakers > 0:
+            g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
+        else:
+            g = None
+
+        logw = self.sdp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w) * sdp_ratio \
+            + self.dp(x, x_mask, g=g) * (1 - sdp_ratio)
+
+        w = torch.exp(logw) * x_mask * length_scale
+        w_ceil = torch.ceil(w)
+        y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
+        y_mask = torch.unsqueeze(commons.sequence_mask(y_lengths, None), 1).to(x_mask.dtype)
+        attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+        attn = commons.generate_path(w_ceil, attn_mask)
+
+        m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
+        logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
+
+        z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
+        z = self.flow(z_p, y_mask, g=g, reverse=True)
+        o = self.dec((z * y_mask)[:,:,:max_len], g=g)
+        return o, attn, y_mask, (z, z_p, m_p, logs_p)
+
+    def voice_conversion(self, y, y_lengths, sid_src, sid_tgt, tau=1.0):
+        g_src = sid_src
+        g_tgt = sid_tgt
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src if not self.zero_g else torch.zeros_like(g_src), tau=tau)
+        z_p = self.flow(z, y_mask, g=g_src)
+        z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True)
+        o_hat = self.dec(z_hat * y_mask, g=g_tgt if not self.zero_g else torch.zeros_like(g_tgt))
+        return o_hat, y_mask, (z, z_p, z_hat)

utils/tts/openvoice/modules.py

@@ -0,0 +1,598 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+from utils.tts.openvoice import commons
+from utils.tts.openvoice.commons import init_weights, get_padding
+from utils.tts.openvoice.transforms import piecewise_rational_quadratic_transform
+from utils.tts.openvoice.attentions import Encoder
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        hidden_channels,
+        out_channels,
+        kernel_size,
+        n_layers,
+        p_dropout,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(
+                in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
+            )
+        )
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(
+                    hidden_channels,
+                    hidden_channels,
+                    kernel_size,
+                    padding=kernel_size // 2,
+                )
+            )
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+    Dilated and Depth-Separable Convolution
+    """
+
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+
+        self.drop = nn.Dropout(p_dropout)
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(
+                    channels,
+                    channels,
+                    kernel_size,
+                    groups=channels,
+                    dilation=dilation,
+                    padding=padding,
+                )
+            )
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g=None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+        p_dropout=0,
+    ):
+        super(WN, self).__init__()
+        assert kernel_size % 2 == 1
+        self.hidden_channels = hidden_channels
+        self.kernel_size = (kernel_size,)
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = p_dropout
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(p_dropout)
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(
+                gin_channels, 2 * hidden_channels * n_layers, 1
+            )
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(
+                hidden_channels,
+                2 * hidden_channels,
+                kernel_size,
+                dilation=dilation,
+                padding=padding,
+            )
+            in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(self, x, x_mask, g=None, **kwargs):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i in range(self.n_layers):
+            x_in = self.in_layers[i](x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = self.res_skip_layers[i](acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, : self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels :, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[2],
+                        padding=get_padding(kernel_size, dilation[2]),
+                    )
+                ),
+            ]
+        )
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+            ]
+        )
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+            ]
+        )
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    def forward(self, x, *args, reverse=False, **kwargs):
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        p_dropout=0,
+        gin_channels=0,
+        mean_only=False,
+    ):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            p_dropout=p_dropout,
+            gin_channels=gin_channels,
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+
+class ConvFlow(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        n_layers,
+        num_bins=10,
+        tail_bound=5.0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
+        self.proj = nn.Conv1d(
+            filter_channels, self.half_channels * (num_bins * 3 - 1), 1
+        )
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
+        unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
+            self.filter_channels
+        )
+        unnormalized_derivatives = h[..., 2 * self.num_bins :]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x
+
+
+class TransformerCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        n_layers,
+        n_heads,
+        p_dropout=0,
+        filter_channels=0,
+        mean_only=False,
+        wn_sharing_parameter=None,
+        gin_channels=0,
+    ):
+        assert n_layers == 3, n_layers
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = (
+            Encoder(
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+                isflow=True,
+                gin_channels=gin_channels,
+            )
+            if wn_sharing_parameter is None
+            else wn_sharing_parameter
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x

utils/tts/openvoice/openvoice_app.py

@@ -0,0 +1,275 @@
+import os
+import torch
+import argparse
+import gradio as gr
+from zipfile import ZipFile
+import langid
+from openvoice import se_extractor
+from openvoice.api import BaseSpeakerTTS, ToneColorConverter
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--share", action='store_true', default=False, help="make link public")
+args = parser.parse_args()
+
+en_ckpt_base = 'checkpoints/base_speakers/EN'
+zh_ckpt_base = 'checkpoints/base_speakers/ZH'
+ckpt_converter = 'checkpoints/converter'
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+output_dir = 'outputs'
+os.makedirs(output_dir, exist_ok=True)
+
+# load models
+en_base_speaker_tts = BaseSpeakerTTS(f'{en_ckpt_base}/config.json', device=device)
+en_base_speaker_tts.load_ckpt(f'{en_ckpt_base}/checkpoint.pth')
+zh_base_speaker_tts = BaseSpeakerTTS(f'{zh_ckpt_base}/config.json', device=device)
+zh_base_speaker_tts.load_ckpt(f'{zh_ckpt_base}/checkpoint.pth')
+tone_color_converter = ToneColorConverter(f'{ckpt_converter}/config.json', device=device)
+tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')
+
+# load speaker embeddings
+en_source_default_se = torch.load(f'{en_ckpt_base}/en_default_se.pth').to(device)
+en_source_style_se = torch.load(f'{en_ckpt_base}/en_style_se.pth').to(device)
+zh_source_se = torch.load(f'{zh_ckpt_base}/zh_default_se.pth').to(device)
+
+# This online demo mainly supports English and Chinese
+supported_languages = ['zh', 'en']
+
+def predict(prompt, style, audio_file_pth, agree):
+    # initialize a empty info
+    text_hint = ''
+    # agree with the terms
+    if agree == False:
+        text_hint += '[ERROR] Please accept the Terms & Condition!\n'
+        gr.Warning("Please accept the Terms & Condition!")
+        return (
+            text_hint,
+            None,
+            None,
+        )
+
+    # first detect the input language
+    language_predicted = langid.classify(prompt)[0].strip()  
+    print(f"Detected language:{language_predicted}")
+
+    if language_predicted not in supported_languages:
+        text_hint += f"[ERROR] The detected language {language_predicted} for your input text is not in our Supported Languages: {supported_languages}\n"
+        gr.Warning(
+            f"The detected language {language_predicted} for your input text is not in our Supported Languages: {supported_languages}"
+        )
+
+        return (
+            text_hint,
+            None,
+            None,
+        )
+    
+    if language_predicted == "zh":
+        tts_model = zh_base_speaker_tts
+        source_se = zh_source_se
+        language = 'Chinese'
+        if style not in ['default']:
+            text_hint += f"[ERROR] The style {style} is not supported for Chinese, which should be in ['default']\n"
+            gr.Warning(f"The style {style} is not supported for Chinese, which should be in ['default']")
+            return (
+                text_hint,
+                None,
+                None,
+            )
+
+    else:
+        tts_model = en_base_speaker_tts
+        if style == 'default':
+            source_se = en_source_default_se
+        else:
+            source_se = en_source_style_se
+        language = 'English'
+        if style not in ['default', 'whispering', 'shouting', 'excited', 'cheerful', 'terrified', 'angry', 'sad', 'friendly']:
+            text_hint += f"[ERROR] The style {style} is not supported for English, which should be in ['default', 'whispering', 'shouting', 'excited', 'cheerful', 'terrified', 'angry', 'sad', 'friendly']\n"
+            gr.Warning(f"The style {style} is not supported for English, which should be in ['default', 'whispering', 'shouting', 'excited', 'cheerful', 'terrified', 'angry', 'sad', 'friendly']")
+            return (
+                text_hint,
+                None,
+                None,
+            )
+
+    speaker_wav = audio_file_pth
+
+    if len(prompt) < 2:
+        text_hint += f"[ERROR] Please give a longer prompt text \n"
+        gr.Warning("Please give a longer prompt text")
+        return (
+            text_hint,
+            None,
+            None,
+        )
+    if len(prompt) > 200:
+        text_hint += f"[ERROR] Text length limited to 200 characters for this demo, please try shorter text. You can clone our open-source repo and try for your usage \n"
+        gr.Warning(
+            "Text length limited to 200 characters for this demo, please try shorter text. You can clone our open-source repo for your usage"
+        )
+        return (
+            text_hint,
+            None,
+            None,
+        )
+    
+    # note diffusion_conditioning not used on hifigan (default mode), it will be empty but need to pass it to model.inference
+    try:
+        target_se, audio_name = se_extractor.get_se(speaker_wav, tone_color_converter, target_dir='processed', vad=True)
+    except Exception as e:
+        text_hint += f"[ERROR] Get target tone color error {str(e)} \n"
+        gr.Warning(
+            "[ERROR] Get target tone color error {str(e)} \n"
+        )
+        return (
+            text_hint,
+            None,
+            None,
+        )
+
+    src_path = f'{output_dir}/tmp.wav'
+    tts_model.tts(prompt, src_path, speaker=style, language=language)
+
+    save_path = f'{output_dir}/output.wav'
+    # Run the tone color converter
+    encode_message = "@MyShell"
+    tone_color_converter.convert(
+        audio_src_path=src_path, 
+        src_se=source_se, 
+        tgt_se=target_se, 
+        output_path=save_path,
+        message=encode_message)
+
+    text_hint += f'''Get response successfully \n'''
+
+    return (
+        text_hint,
+        save_path,
+        speaker_wav,
+    )
+
+
+
+title = "MyShell OpenVoice"
+
+description = """
+We introduce OpenVoice, a versatile instant voice cloning approach that requires only a short audio clip from the reference speaker to replicate their voice and generate speech in multiple languages. OpenVoice enables granular control over voice styles, including emotion, accent, rhythm, pauses, and intonation, in addition to replicating the tone color of the reference speaker. OpenVoice also achieves zero-shot cross-lingual voice cloning for languages not included in the massive-speaker training set.
+"""
+
+markdown_table = """
+<div align="center" style="margin-bottom: 10px;">
+
+|               |               |               |
+| :-----------: | :-----------: | :-----------: | 
+| **OpenSource Repo** | **Project Page** | **Join the Community** |        
+| <div style='text-align: center;'><a style="display:inline-block,align:center" href='https://github.com/myshell-ai/OpenVoice'><img src='https://img.shields.io/github/stars/myshell-ai/OpenVoice?style=social' /></a></div> | [OpenVoice](https://research.myshell.ai/open-voice) | [![Discord](https://img.shields.io/discord/1122227993805336617?color=%239B59B6&label=%20Discord%20)](https://discord.gg/myshell) |
+
+</div>
+"""
+
+markdown_table_v2 = """
+<div align="center" style="margin-bottom: 2px;">
+
+|               |               |               |              |
+| :-----------: | :-----------: | :-----------: | :-----------: | 
+| **OpenSource Repo** | <div style='text-align: center;'><a style="display:inline-block,align:center" href='https://github.com/myshell-ai/OpenVoice'><img src='https://img.shields.io/github/stars/myshell-ai/OpenVoice?style=social' /></a></div> |  **Project Page** |  [OpenVoice](https://research.myshell.ai/open-voice) |     
+
+| | |
+| :-----------: | :-----------: |
+**Join the Community** |   [![Discord](https://img.shields.io/discord/1122227993805336617?color=%239B59B6&label=%20Discord%20)](https://discord.gg/myshell) |
+
+</div>
+"""
+content = """
+<div>
+  <strong>If the generated voice does not sound like the reference voice, please refer to <a href='https://github.com/myshell-ai/OpenVoice/blob/main/docs/QA.md'>this QnA</a>.</strong> <strong>For multi-lingual & cross-lingual examples, please refer to <a href='https://github.com/myshell-ai/OpenVoice/blob/main/demo_part2.ipynb'>this jupyter notebook</a>.</strong>
+  This online demo mainly supports <strong>English</strong>. The <em>default</em> style also supports <strong>Chinese</strong>. But OpenVoice can adapt to any other language as long as a base speaker is provided.
+</div>
+"""
+wrapped_markdown_content = f"<div style='border: 1px solid #000; padding: 10px;'>{content}</div>"
+
+
+examples = [
+    [
+        "今天天气真好，我们一起出去吃饭吧。",
+        'default',
+        "resources/demo_speaker1.mp3",
+        True,
+    ],[
+        "This audio is generated by open voice with a half-performance model.",
+        'whispering',
+        "resources/demo_speaker2.mp3",
+        True,
+    ],
+    [
+        "He hoped there would be stew for dinner, turnips and carrots and bruised potatoes and fat mutton pieces to be ladled out in thick, peppered, flour-fattened sauce.",
+        'sad',
+        "resources/demo_speaker0.mp3",
+        True,
+    ],
+]
+
+with gr.Blocks(analytics_enabled=False) as demo:
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                gr.Markdown(
+                    """
+                    ## <img src="https://huggingface.co/spaces/myshell-ai/OpenVoice/raw/main/logo.jpg" height="40"/>
+                    """
+                )
+            with gr.Row():    
+                gr.Markdown(markdown_table_v2)
+            with gr.Row():
+                gr.Markdown(description)
+        with gr.Column():
+            gr.Video('https://github.com/myshell-ai/OpenVoice/assets/40556743/3cba936f-82bf-476c-9e52-09f0f417bb2f', autoplay=True)
+            
+    with gr.Row():
+        gr.HTML(wrapped_markdown_content)
+
+    with gr.Row():
+        with gr.Column():
+            input_text_gr = gr.Textbox(
+                label="Text Prompt",
+                info="One or two sentences at a time is better. Up to 200 text characters.",
+                value="He hoped there would be stew for dinner, turnips and carrots and bruised potatoes and fat mutton pieces to be ladled out in thick, peppered, flour-fattened sauce.",
+            )
+            style_gr = gr.Dropdown(
+                label="Style",
+                info="Select a style of output audio for the synthesised speech. (Chinese only support 'default' now)",
+                choices=['default', 'whispering', 'cheerful', 'terrified', 'angry', 'sad', 'friendly'],
+                max_choices=1,
+                value="default",
+            )
+            ref_gr = gr.Audio(
+                label="Reference Audio",
+                info="Click on the ✎ button to upload your own target speaker audio",
+                type="filepath",
+                value="resources/demo_speaker2.mp3",
+            )
+            tos_gr = gr.Checkbox(
+                label="Agree",
+                value=False,
+                info="I agree to the terms of the cc-by-nc-4.0 license-: https://github.com/myshell-ai/OpenVoice/blob/main/LICENSE",
+            )
+
+            tts_button = gr.Button("Send", elem_id="send-btn", visible=True)
+
+
+        with gr.Column():
+            out_text_gr = gr.Text(label="Info")
+            audio_gr = gr.Audio(label="Synthesised Audio", autoplay=True)
+            ref_audio_gr = gr.Audio(label="Reference Audio Used")
+
+            gr.Examples(examples,
+                        label="Examples",
+                        inputs=[input_text_gr, style_gr, ref_gr, tos_gr],
+                        outputs=[out_text_gr, audio_gr, ref_audio_gr],
+                        fn=predict,
+                        cache_examples=False,)
+            tts_button.click(predict, [input_text_gr, style_gr, ref_gr, tos_gr], outputs=[out_text_gr, audio_gr, ref_audio_gr])
+
+demo.queue()  
+demo.launch(debug=True, show_api=True, share=args.share)

utils/tts/openvoice/se_extractor.py

@@ -0,0 +1,152 @@
+import os
+import glob
+import torch
+import hashlib
+import librosa
+import base64
+from glob import glob
+import numpy as np
+from pydub import AudioSegment
+from faster_whisper import WhisperModel
+import hashlib
+import base64
+import librosa
+from whisper_timestamped.transcribe import get_audio_tensor, get_vad_segments
+
+model_size = "medium"
+# Run on GPU with FP16
+model = None
+def split_audio_whisper(audio_path, audio_name, target_dir='processed'):
+    global model
+    if model is None:
+        model = WhisperModel(model_size, device="cuda", compute_type="float16")
+    audio = AudioSegment.from_file(audio_path)
+    max_len = len(audio)
+
+    target_folder = os.path.join(target_dir, audio_name)
+    
+    segments, info = model.transcribe(audio_path, beam_size=5, word_timestamps=True)
+    segments = list(segments)    
+
+    # create directory
+    os.makedirs(target_folder, exist_ok=True)
+    wavs_folder = os.path.join(target_folder, 'wavs')
+    os.makedirs(wavs_folder, exist_ok=True)
+
+    # segments
+    s_ind = 0
+    start_time = None
+    
+    for k, w in enumerate(segments):
+        # process with the time
+        if k == 0:
+            start_time = max(0, w.start)
+
+        end_time = w.end
+
+        # calculate confidence
+        if len(w.words) > 0:
+            confidence = sum([s.probability for s in w.words]) / len(w.words)
+        else:
+            confidence = 0.
+        # clean text
+        text = w.text.replace('...', '')
+
+        # left 0.08s for each audios
+        audio_seg = audio[int( start_time * 1000) : min(max_len, int(end_time * 1000) + 80)]
+
+        # segment file name
+        fname = f"{audio_name}_seg{s_ind}.wav"
+
+        # filter out the segment shorter than 1.5s and longer than 20s
+        save = audio_seg.duration_seconds > 1.5 and \
+                audio_seg.duration_seconds < 20. and \
+                len(text) >= 2 and len(text) < 200 
+
+        if save:
+            output_file = os.path.join(wavs_folder, fname)
+            audio_seg.export(output_file, format='wav')
+
+        if k < len(segments) - 1:
+            start_time = max(0, segments[k+1].start - 0.08)
+
+        s_ind = s_ind + 1
+    return wavs_folder
+
+
+def split_audio_vad(audio_path, audio_name, target_dir, split_seconds=10.0):
+    SAMPLE_RATE = 16000
+    audio_vad = get_audio_tensor(audio_path)
+    segments = get_vad_segments(
+        audio_vad,
+        output_sample=True,
+        min_speech_duration=0.1,
+        min_silence_duration=1,
+        method="silero",
+    )
+    segments = [(seg["start"], seg["end"]) for seg in segments]
+    segments = [(float(s) / SAMPLE_RATE, float(e) / SAMPLE_RATE) for s,e in segments]
+    print(segments)
+    audio_active = AudioSegment.silent(duration=0)
+    audio = AudioSegment.from_file(audio_path)
+
+    for start_time, end_time in segments:
+        audio_active += audio[int( start_time * 1000) : int(end_time * 1000)]
+    
+    audio_dur = audio_active.duration_seconds
+    print(f'after vad: dur = {audio_dur}')
+    target_folder = os.path.join(target_dir, audio_name)
+    wavs_folder = os.path.join(target_folder, 'wavs')
+    os.makedirs(wavs_folder, exist_ok=True)
+    start_time = 0.
+    count = 0
+    num_splits = int(np.round(audio_dur / split_seconds))
+    assert num_splits > 0, 'input audio is too short'
+    interval = audio_dur / num_splits
+
+    for i in range(num_splits):
+        end_time = min(start_time + interval, audio_dur)
+        if i == num_splits - 1:
+            end_time = audio_dur
+        output_file = f"{wavs_folder}/{audio_name}_seg{count}.wav"
+        audio_seg = audio_active[int(start_time * 1000): int(end_time * 1000)]
+        audio_seg.export(output_file, format='wav')
+        start_time = end_time
+        count += 1
+    return wavs_folder
+
+def hash_numpy_array(audio_path):
+    array, _ = librosa.load(audio_path, sr=None, mono=True)
+    # Convert the array to bytes
+    array_bytes = array.tobytes()
+    # Calculate the hash of the array bytes
+    hash_object = hashlib.sha256(array_bytes)
+    hash_value = hash_object.digest()
+    # Convert the hash value to base64
+    base64_value = base64.b64encode(hash_value)
+    return base64_value.decode('utf-8')[:16].replace('/', '_^')
+
+def get_se(audio_path, vc_model, target_dir='processed', vad=True):
+    device = vc_model.device
+    version = vc_model.version
+    print("OpenVoice version:", version)
+
+    audio_name = f"{os.path.basename(audio_path).rsplit('.', 1)[0]}_{version}_{hash_numpy_array(audio_path)}"
+    se_path = os.path.join(target_dir, audio_name, 'se.pth')
+
+    # if os.path.isfile(se_path):
+    #     se = torch.load(se_path).to(device)
+    #     return se, audio_name
+    # if os.path.isdir(audio_path):
+    #     wavs_folder = audio_path
+    
+    if vad:
+        wavs_folder = split_audio_vad(audio_path, target_dir=target_dir, audio_name=audio_name)
+    else:
+        wavs_folder = split_audio_whisper(audio_path, target_dir=target_dir, audio_name=audio_name)
+    
+    audio_segs = glob(f'{wavs_folder}/*.wav')
+    if len(audio_segs) == 0:
+        raise NotImplementedError('No audio segments found!')
+    
+    return vc_model.extract_se(audio_segs, se_save_path=se_path), audio_name

utils/tts/openvoice/text/__init__.py

@@ -0,0 +1,79 @@
+""" from https://github.com/keithito/tacotron """
+from utils.tts.openvoice.text import cleaners
+from utils.tts.openvoice.text.symbols import symbols
+
+
+# Mappings from symbol to numeric ID and vice versa:
+_symbol_to_id = {s: i for i, s in enumerate(symbols)}
+_id_to_symbol = {i: s for i, s in enumerate(symbols)}
+
+
+def text_to_sequence(text, symbols, cleaner_names):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+      cleaner_names: names of the cleaner functions to run the text through
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  sequence = []
+  symbol_to_id = {s: i for i, s in enumerate(symbols)}
+  clean_text = _clean_text(text, cleaner_names)
+  print(clean_text)
+  print(f" length:{len(clean_text)}")
+  for symbol in clean_text:
+    if symbol not in symbol_to_id.keys():
+      continue
+    symbol_id = symbol_to_id[symbol]
+    sequence += [symbol_id]
+  print(f" length:{len(sequence)}")
+  return sequence
+
+
+def cleaned_text_to_sequence(cleaned_text, symbols):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  symbol_to_id = {s: i for i, s in enumerate(symbols)}
+  sequence = [symbol_to_id[symbol] for symbol in cleaned_text if symbol in symbol_to_id.keys()]
+  return sequence
+
+
+
+from utils.tts.openvoice.text.symbols import language_tone_start_map
+def cleaned_text_to_sequence_vits2(cleaned_text, tones, language, symbols, languages):
+    """Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+    """
+    symbol_to_id = {s: i for i, s in enumerate(symbols)}
+    language_id_map = {s: i for i, s in enumerate(languages)}
+    phones = [symbol_to_id[symbol] for symbol in cleaned_text]
+    tone_start = language_tone_start_map[language]
+    tones = [i + tone_start for i in tones]
+    lang_id = language_id_map[language]
+    lang_ids = [lang_id for i in phones]
+    return phones, tones, lang_ids
+
+
+def sequence_to_text(sequence):
+  '''Converts a sequence of IDs back to a string'''
+  result = ''
+  for symbol_id in sequence:
+    s = _id_to_symbol[symbol_id]
+    result += s
+  return result
+
+
+def _clean_text(text, cleaner_names):
+  for name in cleaner_names:
+    cleaner = getattr(cleaners, name)
+    if not cleaner:
+      raise Exception('Unknown cleaner: %s' % name)
+    text = cleaner(text)
+  return text

utils/tts/openvoice/text/cleaners.py

@@ -0,0 +1,16 @@
+import re
+from utils.tts.openvoice.text.english import english_to_lazy_ipa, english_to_ipa2, english_to_lazy_ipa2
+from utils.tts.openvoice.text.mandarin import number_to_chinese, chinese_to_bopomofo, latin_to_bopomofo, chinese_to_romaji, chinese_to_lazy_ipa, chinese_to_ipa, chinese_to_ipa2
+
+def cjke_cleaners2(text):
+    text = re.sub(r'\[ZH\](.*?)\[ZH\]',
+                  lambda x: chinese_to_ipa(x.group(1))+' ', text)
+    text = re.sub(r'\[JA\](.*?)\[JA\]',
+                  lambda x: japanese_to_ipa2(x.group(1))+' ', text)
+    text = re.sub(r'\[KO\](.*?)\[KO\]',
+                  lambda x: korean_to_ipa(x.group(1))+' ', text)
+    text = re.sub(r'\[EN\](.*?)\[EN\]',
+                  lambda x: english_to_ipa2(x.group(1))+' ', text)
+    text = re.sub(r'\s+$', '', text)
+    text = re.sub(r'([^\.,!\?\-…~])$', r'\1.', text)
+    return text

utils/tts/openvoice/text/english.py

@@ -0,0 +1,188 @@
+""" from https://github.com/keithito/tacotron """
+
+'''
+Cleaners are transformations that run over the input text at both training and eval time.
+
+Cleaners can be selected by passing a comma-delimited list of cleaner names as the "cleaners"
+hyperparameter. Some cleaners are English-specific. You'll typically want to use:
+  1. "english_cleaners" for English text
+  2. "transliteration_cleaners" for non-English text that can be transliterated to ASCII using
+     the Unidecode library (https://pypi.python.org/pypi/Unidecode)
+  3. "basic_cleaners" if you do not want to transliterate (in this case, you should also update
+     the symbols in symbols.py to match your data).
+'''
+
+
+# Regular expression matching whitespace:
+
+
+import re
+import inflect
+from unidecode import unidecode
+import eng_to_ipa as ipa
+_inflect = inflect.engine()
+_comma_number_re = re.compile(r'([0-9][0-9\,]+[0-9])')
+_decimal_number_re = re.compile(r'([0-9]+\.[0-9]+)')
+_pounds_re = re.compile(r'£([0-9\,]*[0-9]+)')
+_dollars_re = re.compile(r'\$([0-9\.\,]*[0-9]+)')
+_ordinal_re = re.compile(r'[0-9]+(st|nd|rd|th)')
+_number_re = re.compile(r'[0-9]+')
+
+# List of (regular expression, replacement) pairs for abbreviations:
+_abbreviations = [(re.compile('\\b%s\\.' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('mrs', 'misess'),
+    ('mr', 'mister'),
+    ('dr', 'doctor'),
+    ('st', 'saint'),
+    ('co', 'company'),
+    ('jr', 'junior'),
+    ('maj', 'major'),
+    ('gen', 'general'),
+    ('drs', 'doctors'),
+    ('rev', 'reverend'),
+    ('lt', 'lieutenant'),
+    ('hon', 'honorable'),
+    ('sgt', 'sergeant'),
+    ('capt', 'captain'),
+    ('esq', 'esquire'),
+    ('ltd', 'limited'),
+    ('col', 'colonel'),
+    ('ft', 'fort'),
+]]
+
+
+# List of (ipa, lazy ipa) pairs:
+_lazy_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('æ', 'e'),
+    ('ɑ', 'a'),
+    ('ɔ', 'o'),
+    ('ð', 'z'),
+    ('θ', 's'),
+    ('ɛ', 'e'),
+    ('ɪ', 'i'),
+    ('ʊ', 'u'),
+    ('ʒ', 'ʥ'),
+    ('ʤ', 'ʥ'),
+    ('ˈ', '↓'),
+]]
+
+# List of (ipa, lazy ipa2) pairs:
+_lazy_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('ð', 'z'),
+    ('θ', 's'),
+    ('ʒ', 'ʑ'),
+    ('ʤ', 'dʑ'),
+    ('ˈ', '↓'),
+]]
+
+# List of (ipa, ipa2) pairs
+_ipa_to_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('r', 'ɹ'),
+    ('ʤ', 'dʒ'),
+    ('ʧ', 'tʃ')
+]]
+
+
+def expand_abbreviations(text):
+    for regex, replacement in _abbreviations:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def collapse_whitespace(text):
+    return re.sub(r'\s+', ' ', text)
+
+
+def _remove_commas(m):
+    return m.group(1).replace(',', '')
+
+
+def _expand_decimal_point(m):
+    return m.group(1).replace('.', ' point ')
+
+
+def _expand_dollars(m):
+    match = m.group(1)
+    parts = match.split('.')
+    if len(parts) > 2:
+        return match + ' dollars'  # Unexpected format
+    dollars = int(parts[0]) if parts[0] else 0
+    cents = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    if dollars and cents:
+        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
+        cent_unit = 'cent' if cents == 1 else 'cents'
+        return '%s %s, %s %s' % (dollars, dollar_unit, cents, cent_unit)
+    elif dollars:
+        dollar_unit = 'dollar' if dollars == 1 else 'dollars'
+        return '%s %s' % (dollars, dollar_unit)
+    elif cents:
+        cent_unit = 'cent' if cents == 1 else 'cents'
+        return '%s %s' % (cents, cent_unit)
+    else:
+        return 'zero dollars'
+
+
+def _expand_ordinal(m):
+    return _inflect.number_to_words(m.group(0))
+
+
+def _expand_number(m):
+    num = int(m.group(0))
+    if num > 1000 and num < 3000:
+        if num == 2000:
+            return 'two thousand'
+        elif num > 2000 and num < 2010:
+            return 'two thousand ' + _inflect.number_to_words(num % 100)
+        elif num % 100 == 0:
+            return _inflect.number_to_words(num // 100) + ' hundred'
+        else:
+            return _inflect.number_to_words(num, andword='', zero='oh', group=2).replace(', ', ' ')
+    else:
+        return _inflect.number_to_words(num, andword='')
+
+
+def normalize_numbers(text):
+    text = re.sub(_comma_number_re, _remove_commas, text)
+    text = re.sub(_pounds_re, r'\1 pounds', text)
+    text = re.sub(_dollars_re, _expand_dollars, text)
+    text = re.sub(_decimal_number_re, _expand_decimal_point, text)
+    text = re.sub(_ordinal_re, _expand_ordinal, text)
+    text = re.sub(_number_re, _expand_number, text)
+    return text
+
+
+def mark_dark_l(text):
+    return re.sub(r'l([^aeiouæɑɔəɛɪʊ ]*(?: |$))', lambda x: 'ɫ'+x.group(1), text)
+
+
+def english_to_ipa(text):
+    text = unidecode(text).lower()
+    text = expand_abbreviations(text)
+    text = normalize_numbers(text)
+    phonemes = ipa.convert(text)
+    phonemes = collapse_whitespace(phonemes)
+    return phonemes
+
+
+def english_to_lazy_ipa(text):
+    text = english_to_ipa(text)
+    for regex, replacement in _lazy_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def english_to_ipa2(text):
+    text = english_to_ipa(text)
+    text = mark_dark_l(text)
+    for regex, replacement in _ipa_to_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text.replace('...', '…')
+
+
+def english_to_lazy_ipa2(text):
+    text = english_to_ipa(text)
+    for regex, replacement in _lazy_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text

utils/tts/openvoice/text/mandarin.py

@@ -0,0 +1,326 @@
+import os
+import sys
+import re
+from pypinyin import lazy_pinyin, BOPOMOFO
+import jieba
+import cn2an
+import logging
+
+
+# List of (Latin alphabet, bopomofo) pairs:
+_latin_to_bopomofo = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('a', 'ㄟˉ'),
+    ('b', 'ㄅㄧˋ'),
+    ('c', 'ㄙㄧˉ'),
+    ('d', 'ㄉㄧˋ'),
+    ('e', 'ㄧˋ'),
+    ('f', 'ㄝˊㄈㄨˋ'),
+    ('g', 'ㄐㄧˋ'),
+    ('h', 'ㄝˇㄑㄩˋ'),
+    ('i', 'ㄞˋ'),
+    ('j', 'ㄐㄟˋ'),
+    ('k', 'ㄎㄟˋ'),
+    ('l', 'ㄝˊㄛˋ'),
+    ('m', 'ㄝˊㄇㄨˋ'),
+    ('n', 'ㄣˉ'),
+    ('o', 'ㄡˉ'),
+    ('p', 'ㄆㄧˉ'),
+    ('q', 'ㄎㄧㄡˉ'),
+    ('r', 'ㄚˋ'),
+    ('s', 'ㄝˊㄙˋ'),
+    ('t', 'ㄊㄧˋ'),
+    ('u', 'ㄧㄡˉ'),
+    ('v', 'ㄨㄧˉ'),
+    ('w', 'ㄉㄚˋㄅㄨˋㄌㄧㄡˋ'),
+    ('x', 'ㄝˉㄎㄨˋㄙˋ'),
+    ('y', 'ㄨㄞˋ'),
+    ('z', 'ㄗㄟˋ')
+]]
+
+# List of (bopomofo, romaji) pairs:
+_bopomofo_to_romaji = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'p⁼wo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p⁼'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't⁼'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k⁼'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'h'),
+    ('ㄐ', 'ʧ⁼'),
+    ('ㄑ', 'ʧʰ'),
+    ('ㄒ', 'ʃ'),
+    ('ㄓ', 'ʦ`⁼'),
+    ('ㄔ', 'ʦ`ʰ'),
+    ('ㄕ', 's`'),
+    ('ㄖ', 'ɹ`'),
+    ('ㄗ', 'ʦ⁼'),
+    ('ㄘ', 'ʦʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ə'),
+    ('ㄝ', 'e'),
+    ('ㄞ', 'ai'),
+    ('ㄟ', 'ei'),
+    ('ㄠ', 'au'),
+    ('ㄡ', 'ou'),
+    ('ㄧㄢ', 'yeNN'),
+    ('ㄢ', 'aNN'),
+    ('ㄧㄣ', 'iNN'),
+    ('ㄣ', 'əNN'),
+    ('ㄤ', 'aNg'),
+    ('ㄧㄥ', 'iNg'),
+    ('ㄨㄥ', 'uNg'),
+    ('ㄩㄥ', 'yuNg'),
+    ('ㄥ', 'əNg'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'ɥ'),
+    ('ˉ', '→'),
+    ('ˊ', '↑'),
+    ('ˇ', '↓↑'),
+    ('ˋ', '↓'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+# List of (romaji, ipa) pairs:
+_romaji_to_ipa = [(re.compile('%s' % x[0], re.IGNORECASE), x[1]) for x in [
+    ('ʃy', 'ʃ'),
+    ('ʧʰy', 'ʧʰ'),
+    ('ʧ⁼y', 'ʧ⁼'),
+    ('NN', 'n'),
+    ('Ng', 'ŋ'),
+    ('y', 'j'),
+    ('h', 'x')
+]]
+
+# List of (bopomofo, ipa) pairs:
+_bopomofo_to_ipa = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'p⁼wo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p⁼'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't⁼'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k⁼'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'x'),
+    ('ㄐ', 'tʃ⁼'),
+    ('ㄑ', 'tʃʰ'),
+    ('ㄒ', 'ʃ'),
+    ('ㄓ', 'ts`⁼'),
+    ('ㄔ', 'ts`ʰ'),
+    ('ㄕ', 's`'),
+    ('ㄖ', 'ɹ`'),
+    ('ㄗ', 'ts⁼'),
+    ('ㄘ', 'tsʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ə'),
+    ('ㄝ', 'ɛ'),
+    ('ㄞ', 'aɪ'),
+    ('ㄟ', 'eɪ'),
+    ('ㄠ', 'ɑʊ'),
+    ('ㄡ', 'oʊ'),
+    ('ㄧㄢ', 'jɛn'),
+    ('ㄩㄢ', 'ɥæn'),
+    ('ㄢ', 'an'),
+    ('ㄧㄣ', 'in'),
+    ('ㄩㄣ', 'ɥn'),
+    ('ㄣ', 'ən'),
+    ('ㄤ', 'ɑŋ'),
+    ('ㄧㄥ', 'iŋ'),
+    ('ㄨㄥ', 'ʊŋ'),
+    ('ㄩㄥ', 'jʊŋ'),
+    ('ㄥ', 'əŋ'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'ɥ'),
+    ('ˉ', '→'),
+    ('ˊ', '↑'),
+    ('ˇ', '↓↑'),
+    ('ˋ', '↓'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+# List of (bopomofo, ipa2) pairs:
+_bopomofo_to_ipa2 = [(re.compile('%s' % x[0]), x[1]) for x in [
+    ('ㄅㄛ', 'pwo'),
+    ('ㄆㄛ', 'pʰwo'),
+    ('ㄇㄛ', 'mwo'),
+    ('ㄈㄛ', 'fwo'),
+    ('ㄅ', 'p'),
+    ('ㄆ', 'pʰ'),
+    ('ㄇ', 'm'),
+    ('ㄈ', 'f'),
+    ('ㄉ', 't'),
+    ('ㄊ', 'tʰ'),
+    ('ㄋ', 'n'),
+    ('ㄌ', 'l'),
+    ('ㄍ', 'k'),
+    ('ㄎ', 'kʰ'),
+    ('ㄏ', 'h'),
+    ('ㄐ', 'tɕ'),
+    ('ㄑ', 'tɕʰ'),
+    ('ㄒ', 'ɕ'),
+    ('ㄓ', 'tʂ'),
+    ('ㄔ', 'tʂʰ'),
+    ('ㄕ', 'ʂ'),
+    ('ㄖ', 'ɻ'),
+    ('ㄗ', 'ts'),
+    ('ㄘ', 'tsʰ'),
+    ('ㄙ', 's'),
+    ('ㄚ', 'a'),
+    ('ㄛ', 'o'),
+    ('ㄜ', 'ɤ'),
+    ('ㄝ', 'ɛ'),
+    ('ㄞ', 'aɪ'),
+    ('ㄟ', 'eɪ'),
+    ('ㄠ', 'ɑʊ'),
+    ('ㄡ', 'oʊ'),
+    ('ㄧㄢ', 'jɛn'),
+    ('ㄩㄢ', 'yæn'),
+    ('ㄢ', 'an'),
+    ('ㄧㄣ', 'in'),
+    ('ㄩㄣ', 'yn'),
+    ('ㄣ', 'ən'),
+    ('ㄤ', 'ɑŋ'),
+    ('ㄧㄥ', 'iŋ'),
+    ('ㄨㄥ', 'ʊŋ'),
+    ('ㄩㄥ', 'jʊŋ'),
+    ('ㄥ', 'ɤŋ'),
+    ('ㄦ', 'əɻ'),
+    ('ㄧ', 'i'),
+    ('ㄨ', 'u'),
+    ('ㄩ', 'y'),
+    ('ˉ', '˥'),
+    ('ˊ', '˧˥'),
+    ('ˇ', '˨˩˦'),
+    ('ˋ', '˥˩'),
+    ('˙', ''),
+    ('，', ','),
+    ('。', '.'),
+    ('！', '!'),
+    ('？', '?'),
+    ('—', '-')
+]]
+
+
+def number_to_chinese(text):
+    numbers = re.findall(r'\d+(?:\.?\d+)?', text)
+    for number in numbers:
+        text = text.replace(number, cn2an.an2cn(number), 1)
+    return text
+
+
+def chinese_to_bopomofo(text):
+    text = text.replace('、', '，').replace('；', '，').replace('：', '，')
+    words = jieba.lcut(text, cut_all=False)
+    text = ''
+    for word in words:
+        bopomofos = lazy_pinyin(word, BOPOMOFO)
+        if not re.search('[\u4e00-\u9fff]', word):
+            text += word
+            continue
+        for i in range(len(bopomofos)):
+            bopomofos[i] = re.sub(r'([\u3105-\u3129])$', r'\1ˉ', bopomofos[i])
+        if text != '':
+            text += ' '
+        text += ''.join(bopomofos)
+    return text
+
+
+def latin_to_bopomofo(text):
+    for regex, replacement in _latin_to_bopomofo:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_romaji(text):
+    for regex, replacement in _bopomofo_to_romaji:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_ipa(text):
+    for regex, replacement in _bopomofo_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def bopomofo_to_ipa2(text):
+    for regex, replacement in _bopomofo_to_ipa2:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def chinese_to_romaji(text):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_romaji(text)
+    text = re.sub('i([aoe])', r'y\1', text)
+    text = re.sub('u([aoəe])', r'w\1', text)
+    text = re.sub('([ʦsɹ]`[⁼ʰ]?)([→↓↑ ]+|$)',
+                  r'\1ɹ`\2', text).replace('ɻ', 'ɹ`')
+    text = re.sub('([ʦs][⁼ʰ]?)([→↓↑ ]+|$)', r'\1ɹ\2', text)
+    return text
+
+
+def chinese_to_lazy_ipa(text):
+    text = chinese_to_romaji(text)
+    for regex, replacement in _romaji_to_ipa:
+        text = re.sub(regex, replacement, text)
+    return text
+
+
+def chinese_to_ipa(text):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_ipa(text)
+    text = re.sub('i([aoe])', r'j\1', text)
+    text = re.sub('u([aoəe])', r'w\1', text)
+    text = re.sub('([sɹ]`[⁼ʰ]?)([→↓↑ ]+|$)',
+                  r'\1ɹ`\2', text).replace('ɻ', 'ɹ`')
+    text = re.sub('([s][⁼ʰ]?)([→↓↑ ]+|$)', r'\1ɹ\2', text)
+    return text
+
+
+def chinese_to_ipa2(text):
+    text = number_to_chinese(text)
+    text = chinese_to_bopomofo(text)
+    text = latin_to_bopomofo(text)
+    text = bopomofo_to_ipa2(text)
+    text = re.sub(r'i([aoe])', r'j\1', text)
+    text = re.sub(r'u([aoəe])', r'w\1', text)
+    text = re.sub(r'([ʂɹ]ʰ?)([˩˨˧˦˥ ]+|$)', r'\1ʅ\2', text)
+    text = re.sub(r'(sʰ?)([˩˨˧˦˥ ]+|$)', r'\1ɿ\2', text)
+    return text

utils/tts/openvoice/text/symbols.py

@@ -0,0 +1,88 @@
+'''
+Defines the set of symbols used in text input to the model.
+'''
+
+# japanese_cleaners
+# _pad        = '_'
+# _punctuation = ',.!?-'
+# _letters = 'AEINOQUabdefghijkmnoprstuvwyzʃʧ↓↑ '
+
+
+'''# japanese_cleaners2
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'AEINOQUabdefghijkmnoprstuvwyzʃʧʦ↓↑ '
+'''
+
+
+'''# korean_cleaners
+_pad        = '_'
+_punctuation = ',.!?…~'
+_letters = 'ㄱㄴㄷㄹㅁㅂㅅㅇㅈㅊㅋㅌㅍㅎㄲㄸㅃㅆㅉㅏㅓㅗㅜㅡㅣㅐㅔ '
+'''
+
+'''# chinese_cleaners
+_pad        = '_'
+_punctuation = '，。！？—…'
+_letters = 'ㄅㄆㄇㄈㄉㄊㄋㄌㄍㄎㄏㄐㄑㄒㄓㄔㄕㄖㄗㄘㄙㄚㄛㄜㄝㄞㄟㄠㄡㄢㄣㄤㄥㄦㄧㄨㄩˉˊˇˋ˙ '
+'''
+
+# # zh_ja_mixture_cleaners
+# _pad        = '_'
+# _punctuation = ',.!?-~…'
+# _letters = 'AEINOQUabdefghijklmnoprstuvwyzʃʧʦɯɹəɥ⁼ʰ`→↓↑ '
+
+
+'''# sanskrit_cleaners
+_pad        = '_'
+_punctuation = '।'
+_letters = 'ँंःअआइईउऊऋएऐओऔकखगघङचछजझञटठडढणतथदधनपफबभमयरलळवशषसहऽािीुूृॄेैोौ्ॠॢ '
+'''
+
+'''# cjks_cleaners
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'NQabdefghijklmnopstuvwxyzʃʧʥʦɯɹəɥçɸɾβŋɦː⁼ʰ`^#*=→↓↑ '
+'''
+
+'''# thai_cleaners
+_pad        = '_'
+_punctuation = '.!? '
+_letters = 'กขฃคฆงจฉชซฌญฎฏฐฑฒณดตถทธนบปผฝพฟภมยรฤลวศษสหฬอฮฯะัาำิีึืุูเแโใไๅๆ็่้๊๋์'
+'''
+
+# # cjke_cleaners2
+_pad        = '_'
+_punctuation = ',.!?-~…'
+_letters = 'NQabdefghijklmnopstuvwxyzɑæʃʑçɯɪɔɛɹðəɫɥɸʊɾʒθβŋɦ⁼ʰ`^#*=ˈˌ→↓↑ '
+
+
+'''# shanghainese_cleaners
+_pad        = '_'
+_punctuation = ',.!?…'
+_letters = 'abdfghiklmnopstuvyzøŋȵɑɔɕəɤɦɪɿʑʔʰ̩̃ᴀᴇ15678 '
+'''
+
+'''# chinese_dialect_cleaners
+_pad        = '_'
+_punctuation = ',.!?~…─'
+_letters = '#Nabdefghijklmnoprstuvwxyzæçøŋœȵɐɑɒɓɔɕɗɘəɚɛɜɣɤɦɪɭɯɵɷɸɻɾɿʂʅʊʋʌʏʑʔʦʮʰʷˀː˥˦˧˨˩̥̩̃̚ᴀᴇ↑↓∅ⱼ '
+'''
+
+# Export all symbols:
+symbols = [_pad] + list(_punctuation) + list(_letters)
+
+# Special symbol ids
+SPACE_ID = symbols.index(" ")
+
+num_ja_tones = 1
+num_kr_tones = 1
+num_zh_tones = 6
+num_en_tones = 4
+
+language_tone_start_map = {
+    "ZH": 0,
+    "JP": num_zh_tones,
+    "EN": num_zh_tones + num_ja_tones,
+    'KR': num_zh_tones + num_ja_tones + num_en_tones,
+}

utils/tts/openvoice/transforms.py

@@ -0,0 +1,209 @@
+import torch
+from torch.nn import functional as F
+
+import numpy as np
+
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails=None,
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {"tails": tails, "tail_bound": tail_bound}
+
+    outputs, logabsdet = spline_fn(
+        inputs=inputs,
+        unnormalized_widths=unnormalized_widths,
+        unnormalized_heights=unnormalized_heights,
+        unnormalized_derivatives=unnormalized_derivatives,
+        inverse=inverse,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+        **spline_kwargs
+    )
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., -1] += eps
+    return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
+
+
+def unconstrained_rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails="linear",
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == "linear":
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., -1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError("{} tails are not implemented.".format(tails))
+
+    (
+        outputs[inside_interval_mask],
+        logabsdet[inside_interval_mask],
+    ) = rational_quadratic_spline(
+        inputs=inputs[inside_interval_mask],
+        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
+        inverse=inverse,
+        left=-tail_bound,
+        right=tail_bound,
+        bottom=-tail_bound,
+        top=tail_bound,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+    )
+
+    return outputs, logabsdet
+
+
+def rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    left=0.0,
+    right=1.0,
+    bottom=0.0,
+    top=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError("Input to a transform is not within its domain")
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError("Minimal bin width too large for the number of bins")
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError("Minimal bin height too large for the number of bins")
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode="constant", value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., -1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode="constant", value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., -1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        ) + input_heights * (input_delta - input_derivatives)
+        b = input_heights * input_derivatives - (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        )
+        c = -input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * root.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - root).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (
+            input_delta * theta.pow(2) + input_derivatives * theta_one_minus_theta
+        )
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * theta.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - theta).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet

utils/tts/openvoice/utils.py

@@ -0,0 +1,194 @@
+import re
+import json
+import numpy as np
+
+
+def get_hparams_from_file(config_path):
+    with open(config_path, "r", encoding="utf-8") as f:
+        data = f.read()
+    config = json.loads(data)
+
+    hparams = HParams(**config)
+    return hparams
+
+class HParams:
+    def __init__(self, **kwargs):
+        for k, v in kwargs.items():
+            if type(v) == dict:
+                v = HParams(**v)
+            self[k] = v
+
+    def keys(self):
+        return self.__dict__.keys()
+
+    def items(self):
+        return self.__dict__.items()
+
+    def values(self):
+        return self.__dict__.values()
+
+    def __len__(self):
+        return len(self.__dict__)
+
+    def __getitem__(self, key):
+        return getattr(self, key)
+
+    def __setitem__(self, key, value):
+        return setattr(self, key, value)
+
+    def __contains__(self, key):
+        return key in self.__dict__
+
+    def __repr__(self):
+        return self.__dict__.__repr__()
+
+
+def string_to_bits(string, pad_len=8):
+    # Convert each character to its ASCII value
+    ascii_values = [ord(char) for char in string]
+    
+    # Convert ASCII values to binary representation
+    binary_values = [bin(value)[2:].zfill(8) for value in ascii_values]
+    
+    # Convert binary strings to integer arrays
+    bit_arrays = [[int(bit) for bit in binary] for binary in binary_values]
+    
+    # Convert list of arrays to NumPy array
+    numpy_array = np.array(bit_arrays)
+    numpy_array_full = np.zeros((pad_len, 8), dtype=numpy_array.dtype)
+    numpy_array_full[:, 2] = 1
+    max_len = min(pad_len, len(numpy_array))
+    numpy_array_full[:max_len] = numpy_array[:max_len]
+    return numpy_array_full
+
+
+def bits_to_string(bits_array):
+    # Convert each row of the array to a binary string
+    binary_values = [''.join(str(bit) for bit in row) for row in bits_array]
+    
+    # Convert binary strings to ASCII values
+    ascii_values = [int(binary, 2) for binary in binary_values]
+    
+    # Convert ASCII values to characters
+    output_string = ''.join(chr(value) for value in ascii_values)
+    
+    return output_string
+
+
+def split_sentence(text, min_len=10, language_str='[EN]'):
+    if language_str in ['EN']:
+        sentences = split_sentences_latin(text, min_len=min_len)
+    else:
+        sentences = split_sentences_zh(text, min_len=min_len)
+    return sentences
+
+def split_sentences_latin(text, min_len=10):
+    """Split Long sentences into list of short ones
+
+    Args:
+        str: Input sentences.
+
+    Returns:
+        List[str]: list of output sentences.
+    """
+    # deal with dirty sentences
+    text = re.sub('[。！？；]', '.', text)
+    text = re.sub('[，]', ',', text)
+    text = re.sub('[“”]', '"', text)
+    text = re.sub('[‘’]', "'", text)
+    text = re.sub(r"[\<\>\(\)\[\]\"\«\»]+", "", text)
+    text = re.sub('[\n\t ]+', ' ', text)
+    text = re.sub('([,.!?;])', r'\1 $#!', text)
+    # split
+    sentences = [s.strip() for s in text.split('$#!')]
+    if len(sentences[-1]) == 0: del sentences[-1]
+
+    new_sentences = []
+    new_sent = []
+    count_len = 0
+    for ind, sent in enumerate(sentences):
+        # print(sent)
+        new_sent.append(sent)
+        count_len += len(sent.split(" "))
+        if count_len > min_len or ind == len(sentences) - 1:
+            count_len = 0
+            new_sentences.append(' '.join(new_sent))
+            new_sent = []
+    return merge_short_sentences_latin(new_sentences)
+
+
+def merge_short_sentences_latin(sens):
+    """Avoid short sentences by merging them with the following sentence.
+
+    Args:
+        List[str]: list of input sentences.
+
+    Returns:
+        List[str]: list of output sentences.
+    """
+    sens_out = []
+    for s in sens:
+        # If the previous sentence is too short, merge them with
+        # the current sentence.
+        if len(sens_out) > 0 and len(sens_out[-1].split(" ")) <= 2:
+            sens_out[-1] = sens_out[-1] + " " + s
+        else:
+            sens_out.append(s)
+    try:
+        if len(sens_out[-1].split(" ")) <= 2:
+            sens_out[-2] = sens_out[-2] + " " + sens_out[-1]
+            sens_out.pop(-1)
+    except:
+        pass
+    return sens_out
+
+def split_sentences_zh(text, min_len=10):
+    text = re.sub('[。！？；]', '.', text)
+    text = re.sub('[，]', ',', text)
+    # 将文本中的换行符、空格和制表符替换为空格
+    text = re.sub('[\n\t ]+', ' ', text)
+    # 在标点符号后添加一个空格
+    text = re.sub('([,.!?;])', r'\1 $#!', text)
+    # 分隔句子并去除前后空格
+    # sentences = [s.strip() for s in re.split('(。|！|？|；)', text)]
+    sentences = [s.strip() for s in text.split('$#!')]
+    if len(sentences[-1]) == 0: del sentences[-1]
+
+    new_sentences = []
+    new_sent = []
+    count_len = 0
+    for ind, sent in enumerate(sentences):
+        new_sent.append(sent)
+        count_len += len(sent)
+        if count_len > min_len or ind == len(sentences) - 1:
+            count_len = 0
+            new_sentences.append(' '.join(new_sent))
+            new_sent = []
+    return merge_short_sentences_zh(new_sentences)
+
+
+def merge_short_sentences_zh(sens):
+    # return sens
+    """Avoid short sentences by merging them with the following sentence.
+
+    Args:
+        List[str]: list of input sentences.
+
+    Returns:
+        List[str]: list of output sentences.
+    """
+    sens_out = []
+    for s in sens:
+        # If the previous sentense is too short, merge them with
+        # the current sentence.
+        if len(sens_out) > 0 and len(sens_out[-1]) <= 2:
+            sens_out[-1] = sens_out[-1] + " " + s
+        else:
+            sens_out.append(s)
+    try:
+        if len(sens_out[-1]) <= 2:
+            sens_out[-2] = sens_out[-2] + " " + sens_out[-1]
+            sens_out.pop(-1)
+    except:
+        pass
+    return sens_out

utils/tts/openvoice_utils.py

@@ -0,0 +1,352 @@
+import os
+import re
+from glob import glob
+import hashlib
+from tqdm.auto import tqdm
+import soundfile as sf
+import numpy as np
+import torch
+from typing import Optional, Union
+# melo
+from melo.api import TTS
+from melo.utils import get_text_for_tts_infer
+# openvoice
+from .openvoice import se_extractor
+from .openvoice.api import ToneColorConverter
+from .openvoice.mel_processing import spectrogram_torch
+# torchaudio
+import torchaudio.functional as F
+# 存储 BASE SPEAKER 的 embedding(source_se) 的路径
+SOURCE_SE_DIR = r"D:\python\OpenVoice\checkpoints_v2\base_speakers\ses"
+
+# 存储缓存文件的路径
+CACHE_PATH = r"D:\python\OpenVoice\processed"
+
+OPENVOICE_BASE_TTS={
+    "model_type": "open_voice_base_tts",
+    # 转换的语言
+    "language": "ZH",
+}
+
+OPENVOICE_TONE_COLOR_CONVERTER={
+    "model_type": "open_voice_converter",
+    # 模型参数路径
+    "converter_path": r"D:\python\OpenVoice\checkpoints_v2\converter",
+}
+
+class TextToSpeech:
+    def __init__(self,
+                 use_tone_convert=True,
+                 device="cuda",
+                 debug:bool=False,
+                 ):
+        self.debug = debug
+        self.device = device
+        self.use_tone_convert = use_tone_convert
+        # 默认的源说话人 se
+        self.source_se = None
+        # 默认的目标说话人 se
+        self.target_se = None
+
+        self.initialize_base_tts(**OPENVOICE_BASE_TTS)
+        if self.debug:
+            print("use tone converter is", self.use_tone_convert)
+        if self.use_tone_convert:
+            self.initialize_tone_color_converter(**OPENVOICE_TONE_COLOR_CONVERTER)
+            self.initialize_source_se()
+        
+    
+    def initialize_tone_color_converter(self, **kwargs):
+        """
+        初始化 tone color converter
+        """
+        model_type = kwargs.pop('model_type')
+        self.tone_color_converter_model_type = model_type
+        if model_type == 'open_voice_converter':
+            # 加载模型
+            converter_path = kwargs.pop('converter_path')
+            self.tone_color_converter = ToneColorConverter(f'{converter_path}/config.json', self.device)
+            self.tone_color_converter.load_ckpt(f'{converter_path}/checkpoint.pth')
+            if self.debug:
+                print("load tone color converter successfully!")
+        else:
+            raise NotImplementedError(f"only [open_voice_converter] model type expected, but get [{model_type}]. ")
+        
+    def initialize_base_tts(self, **kwargs):
+        """
+        初始化 base tts model
+        """
+        model_type = kwargs.pop('model_type')
+        self.base_tts_model_type = model_type
+        if model_type == "open_voice_base_tts":
+            language = kwargs.pop('language')
+            self.base_tts_model = TTS(language=language, device=self.device)
+            speaker_ids = self.base_tts_model.hps.data.spk2id
+            flag = False
+            for speaker_key in speaker_ids.keys():
+                if flag:
+                    Warning(f'loaded model has more than one speaker, only the first speaker is used. The input speaker ids are {speaker_ids}')
+                    break
+                self.speaker_id = speaker_ids[speaker_key]
+                self.speaker_key = speaker_key.lower().replace('_', '-')
+                flag=True
+            if self.debug:
+                print("load base tts model successfully!")
+            # 第一次使用tts时会加载bert模型
+            self._base_tts("初始化bert模型。")
+        else:
+            raise NotImplementedError(f"only [open_voice_base_tts] model type expected, but get [{model_type}]. ")
+
+    def initialize_source_se(self):
+        """
+        初始化source se
+        """
+        if self.source_se is not None:
+            Warning("replace source speaker embedding with new source speaker embedding!")
+        self.source_se = torch.load(os.path.join(SOURCE_SE_DIR, f"{self.speaker_key}.pth"), map_location=self.device)
+
+    def initialize_target_se(self, se: Union[np.ndarray, torch.Tensor]):
+        """
+        设置 target se
+        param:
+            se: 输入的se，类型可以为np.ndarray或torch.Tensor
+        """
+        if self.target_se is not None:
+            Warning("replace target source speaker embedding with new target speaker embedding!")
+        if isinstance(se, np.ndarray):
+            self.target_se = torch.tensor(se.astype(np.float32)).to(self.device)
+        elif isinstance(se, torch.Tensor):
+            self.target_se = se.float().to(self.device)
+
+    def audio2numpy(self, audio_data: Union[bytes, np.ndarray]):
+        """
+        将字节流的audio转为numpy类型，也可以传入numpy类型
+        return: np.float32
+        """
+        # TODO 是否归一化判断
+        if isinstance(audio_data, bytes):
+            audio_data = np.frombuffer(audio_data, dtype=np.int16).flatten().astype(np.float32) / 32768.0
+        elif isinstance(audio_data, np.ndarray):
+            if audio_data.dtype != np.float32:
+                audio_data = audio_data.astype(np.int16).flatten().astype(np.float32) / 32768.0
+        else:
+            raise TypeError(f"audio_data must be bytes or numpy array, but got {type(audio_data)}")
+        return audio_data
+
+    def audio2emb(self, audio_data: Union[bytes, np.ndarray], rate=44100, vad=True):
+        """
+        将输入的字节流/numpy类型的audio转为speaker embedding
+        param:
+            audio_data: 输入的音频字节
+            rate: 输入音频的采样率
+            vad: 是否使用vad模型
+        return: np.ndarray
+        """
+        audio_data = self.audio2numpy(audio_data)
+
+        from scipy.io import wavfile
+        audio_path = os.path.join(CACHE_PATH, "tmp.wav")
+        wavfile.write(audio_path, rate=rate, data=audio_data)
+
+        se, _ = se_extractor.get_se(audio_path, self.tone_color_converter, target_dir=CACHE_PATH, vad=False)
+        # device = self.tone_color_converter.device
+        # version = self.tone_color_converter.version
+        # if self.debug:
+        #     print("OpenVoice version:", version)
+        
+        # audio_name = f"tmp_{version}_{hashlib.sha256(audio_data.tobytes()).hexdigest()[:16].replace('/','_^')}"
+
+        
+        # if vad:
+        #     wavs_folder = se_extractor.split_audio_vad(audio_path, target_dir=CACHE_PATH, audio_name=audio_name)
+        # else:
+        #     wavs_folder = se_extractor.split_audio_whisper(audio_data, target_dir=CACHE_PATH, audio_name=audio_name)
+        
+        # audio_segs = glob(f'{wavs_folder}/*.wav')
+        # if len(audio_segs) == 0:
+        #     raise NotImplementedError('No audio segments found!')
+        # # se, _ = se_extractor.get_se(audio_data, self.tone_color_converter, CACHE_PATH, vad=False)
+        # se = self.tone_color_converter.extract_se(audio_segs)
+        return se.cpu().detach().numpy()
+    
+    def tensor2numpy(self, audio_data: torch.Tensor):
+        """
+        tensor类型转numpy
+        """
+        return audio_data.cpu().detach().float().numpy()
+    
+    def numpy2bytes(self, audio_data: np.ndarray):
+        """
+        numpy类型转bytes
+        """
+        return (audio_data*32768.0).astype(np.int32).tobytes()
+
+    def _base_tts(self,
+                 text: str,
+                 sdp_ratio=0.2,
+                 noise_scale=0.6,
+                 noise_scale_w=0.8,
+                 speed=1.0,
+                 quite=True):
+        """
+        base语音合成
+        param: 
+            text: 要合成的文本
+            sdp_ratio: SDP在合成时的占比, 理论上此比率越高, 合成的语音语调方差越大.
+            noise_scale: 样本噪声张量的噪声标度。
+            noise_scale_w: 推理中随机持续时间预测器的噪声标度
+            speed: 说话语速
+            quite: 是否显示进度条
+        return: 
+            audio: tensor
+            sr: 生成音频的采样速率
+        """
+        speaker_id = self.speaker_id
+        if self.base_tts_model_type != "open_voice_base_tts":
+            raise NotImplementedError("only [open_voice_base_tts] model type expected.")
+        language = self.base_tts_model.language
+        texts = self.base_tts_model.split_sentences_into_pieces(text, language, quite)
+        audio_list = []
+        if quite:
+            tx = texts
+        else:
+            tx = tqdm(texts)
+        for t in tx:
+            if language in ['EN', 'ZH_MIX_EN']:
+                t = re.sub(r'([a-z])([A-Z])', r'\1 \2', t)
+            device = self.base_tts_model.device
+            bert, ja_bert, phones, tones, lang_ids = get_text_for_tts_infer(t, language, self.base_tts_model.hps, device, self.base_tts_model.symbol_to_id)
+            with torch.no_grad():
+                x_tst = phones.to(device).unsqueeze(0)
+                tones = tones.to(device).unsqueeze(0)
+                lang_ids = lang_ids.to(device).unsqueeze(0)
+                bert = bert.to(device).unsqueeze(0)
+                ja_bert = ja_bert.to(device).unsqueeze(0)
+                x_tst_lengths = torch.LongTensor([phones.size(0)]).to(device)
+                del phones
+                speakers = torch.LongTensor([speaker_id]).to(device)
+                audio = self.base_tts_model.model.infer(
+                    x_tst,
+                    x_tst_lengths,
+                    speakers,
+                    tones,
+                    lang_ids,
+                    bert,
+                    ja_bert,
+                    sdp_ratio=sdp_ratio,
+                    noise_scale=noise_scale,
+                    noise_scale_w = noise_scale_w,
+                    length_scale = 1. / speed,
+                )[0][0, 0].data
+                del x_tst, tones, lang_ids, bert, ja_bert, x_tst_lengths, speakers
+            audio_list.append(audio)
+        torch.cuda.empty_cache()
+        audio_segments = []
+        sr = self.base_tts_model.hps.data.sampling_rate 
+        for segment_data in audio_list:
+            audio_segments.append(segment_data.reshape(-1).contiguous())
+            audio_segments.append(torch.tensor([0]*int((sr * 0.05) / speed), dtype=segment_data.dtype, device=segment_data.device))
+        audio_segments = torch.cat(audio_segments, dim=-1)
+        if self.debug:
+            print("generate base speech!")
+            print("**********************,tts sr",sr)
+            print(f"audio segment length is [{audio_segments.shape}]")
+        return audio_segments, sr
+    
+    def _convert_tone(self,
+                      audio_data: torch.Tensor,
+                      source_se: Optional[np.ndarray]=None,
+                      target_se: Optional[np.ndarray]=None,
+                      tau :float=0.3,
+                      message :str="default"):
+        """
+        音色转换
+        param:
+            audio_data: _base_tts输出的音频数据
+            source_se: 如果为None, 则使用self.source_se
+            target_se: 如果为None, 则使用self.target_se
+            tau:
+            message: 水印信息 TODO
+        return:
+            audio: tensor
+            sr: 生成音频的采样速率
+        """
+        if source_se is None:
+            source_se = self.source_se
+        if target_se is None:
+            target_se = self.target_se
+
+        hps = self.tone_color_converter.hps
+        sr = hps.data.sampling_rate
+        if self.debug:
+            print("**********************************, convert sr", sr)
+        audio_data = audio_data.float()
+
+        with torch.no_grad():
+            y = audio_data.to(self.tone_color_converter.device)
+            y = y.unsqueeze(0)
+            spec = spectrogram_torch(y, hps.data.filter_length,
+                                     sr, hps.data.hop_length, hps.data.win_length,
+                                     center=False).to(self.tone_color_converter.device)
+            spec_lengths = torch.LongTensor([spec.size(-1)]).to(self.tone_color_converter.device)
+            audio = self.tone_color_converter.model.voice_conversion(spec, spec_lengths, sid_src=source_se, sid_tgt=target_se, tau=tau)[0][
+                0, 0].data
+            # audio = self.tone_color_converter.add_watermark(audio, message)
+        if self.debug:
+            print("tone color has been converted!")
+        return audio, sr
+    
+    def tts(self,
+            text: str,
+            sdp_ratio=0.2,
+            noise_scale=0.6,
+            noise_scale_w=0.8,
+            speed=1.0,
+            quite=True,
+            
+            source_se: Optional[np.ndarray]=None,
+            target_se: Optional[np.ndarray]=None,
+            tau :float=0.3,
+            message :str="default"):
+        """
+        整体pipeline
+        _base_tts()
+        _convert_tone()
+        tensor2numpy()
+        numpy2bytes()
+        param:
+            见_base_tts和_convert_tone
+        return:
+            audio: 字节流音频数据
+            sr: 音频数据的采样率
+        """
+        audio, sr = self._base_tts(text, 
+                                   sdp_ratio=sdp_ratio,
+                                   noise_scale=noise_scale,
+                                   noise_scale_w=noise_scale_w,
+                                   speed=speed,
+                                   quite=quite)
+        if self.use_tone_convert:
+            tts_sr = self.base_tts_model.hps.data.sampling_rate
+            converter_sr = self.tone_color_converter.hps.data.sampling_rate
+            audio = F.resample(audio, tts_sr, converter_sr)
+            print(audio.dtype)
+            audio, sr = self._convert_tone(audio,
+                                           source_se=source_se,
+                                           target_se=target_se,
+                                           tau=tau,
+                                           message=message)
+        audio = self.tensor2numpy(audio)
+        audio = self.numpy2bytes(audio)
+        return audio, sr
+    
+    def save_audio(self, audio, sample_rate, save_path):
+        """
+        将numpy类型的音频数据保存至本地
+        param:
+            audio: numpy类型的音频数据
+            sample_rate: 数据采样率
+            save_path: 保存路径
+        """
+        sf.write(save_path, audio, sample_rate)
+        print(f"Audio saved to {save_path}")