MaskGCT,AI语音克隆大模型本地部署(Windows11),基于Python3.11,TTS,文字转语音

前几天，又一款非自回归的文字转语音的AI模型：MaskGCT，开放了源码，和同样非自回归的F5-TTS模型一样，MaskGCT模型也是基于10万小时数据集Emilia训练而来的，精通中英日韩法德6种语言的跨语种合成。数据集Emilia是全球最大且最为多样的高质量多语种语音数据集之一。

本次分享一下如何在本地部署MaskGCT项目，让您的显卡再次发烧。

安装基础依赖

首先确保本地已经安装好Python3.11环境，安装包可以去Python的官方下载:

python.org

随后克隆官方项目:

git clone https://github.com/open-mmlab/Amphion.git

官方提供了基于linux的安装shell脚本：

pip install setuptools ruamel.yaml tqdm   
pip install tensorboard tensorboardX torch==2.0.1  
pip install transformers===4.41.1  
pip install -U encodec  
pip install black==24.1.1  
pip install oss2  
sudo apt-get install espeak-ng  
pip install phonemizer  
pip install g2p_en  
pip install accelerate==0.31.0  
pip install funasr zhconv zhon modelscope  
# pip install git+https://github.com/lhotse-speech/lhotse  
pip install timm  
pip install jieba cn2an  
pip install unidecode  
pip install -U cos-python-sdk-v5  
pip install pypinyin  
pip install jiwer  
pip install omegaconf  
pip install pyworld  
pip install py3langid==0.2.2 LangSegment  
pip install onnxruntime  
pip install pyopenjtalk  
pip install pykakasi  
pip install -U openai-whisper

这里笔者为大家转换为适合Windows的requirements.txt依赖文件：

setuptools   
ruamel.yaml   
tqdm   
transformers===4.41.1  
encodec  
black==24.1.1  
oss2  
phonemizer  
g2p_en  
accelerate==0.31.0  
funasr   
zhconv   
zhon   
modelscope  
timm  
jieba   
cn2an  
unidecode  
cos-python-sdk-v5  
pypinyin  
jiwer  
omegaconf  
pyworld  
py3langid==0.2.2  
LangSegment  
onnxruntime  
pyopenjtalk  
pykakasi  
openai-whisper  
json5

运行命令：

pip3 install -r requirements.txt

安装依赖即可。

安装onnxruntime-gpu:

pip3 install onnxruntime-gpu

安装torch三件套:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 

Windows配置espeak-ng

由于MaskGCT项目后端依赖espeak软件，所以需要在本地进行配置，eSpeak 是一个紧凑的开源文本转语音 (TTS) 合成器，支持多种语言和口音 。它使用“共振峰合成”方法，允许以较小的体积提供多种语言 。语音清晰，可以高速使用，但不如基于人类语音录音的较大合成器自然流畅，而MaskGCT就是在espeak的合成基础上进行二次推理。

首先运行命令安装espeak:

winget install espeak

如果装不上，也可以下载安装包手动安装：

https://sourceforge.net/projects/espeak/files/espeak/espeak-1.48/setup_espeak-1.48.04.exe/download

随后下载espeak-ng安装包：

https://github.com/espeak-ng/espeak-ng/releases

下载后双击安装。

接着把 C:\Program Files\eSpeak NG\libespeak-ng.dll 拷贝到 C:\Program Files (x86)\eSpeak\command_line 目录。

然后把 libespeak-ng.dll 重命名为 espeak-ng.dll

最后把 C:\Program Files (x86)\eSpeak\command_line 目录配置到环境变量即可。

MaskGCT本地推理

都配置好之后，编写推理脚本 local_test.py:

from models.tts.maskgct.maskgct_utils import *  
from huggingface_hub import hf_hub_download  
import safetensors  
import soundfile as sf  
import os  
import argparse  
os.environ['HF_HOME'] = os.path.join(os.path.dirname(__file__), 'hf_download')  
  
print(os.path.join(os.path.dirname(__file__), 'hf_download'))  
  
parser = argparse.ArgumentParser(description="GPT-SoVITS api")  
parser.add_argument("-p", "--prompt_text", type=str, default="说得好像您带我以来我考好过几次一样")  
parser.add_argument("-a", "--audio", type=str, default="./说得好像您带我以来我考好过几次一样.wav")  
parser.add_argument("-t", "--text", type=str, default="你好")  
parser.add_argument("-l", "--language", type=str, default="zh")  
parser.add_argument("-lt", "--target_language", type=str, default="zh")  
args = parser.parse_args()  
  
if __name__ == "__main__":  
  
    # download semantic codec ckpt  
    semantic_code_ckpt = hf_hub_download("amphion/MaskGCT", filename="semantic_codec/model.safetensors")  
  
    # download acoustic codec ckpt  
    codec_encoder_ckpt = hf_hub_download("amphion/MaskGCT", filename="acoustic_codec/model.safetensors")  
    codec_decoder_ckpt = hf_hub_download("amphion/MaskGCT", filename="acoustic_codec/model_1.safetensors")  
  
    # download t2s model ckpt  
    t2s_model_ckpt = hf_hub_download("amphion/MaskGCT", filename="t2s_model/model.safetensors")  
  
    # download s2a model ckpt  
    s2a_1layer_ckpt = hf_hub_download("amphion/MaskGCT", filename="s2a_model/s2a_model_1layer/model.safetensors")  
    s2a_full_ckpt = hf_hub_download("amphion/MaskGCT", filename="s2a_model/s2a_model_full/model.safetensors")  
  
    # build model  
    device = torch.device("cuda")  
    cfg_path = "./models/tts/maskgct/config/maskgct.json"  
    cfg = load_config(cfg_path)  
    # 1. build semantic model (w2v-bert-2.0)  
    semantic_model, semantic_mean, semantic_std = build_semantic_model(device)  
    # 2. build semantic codec  
    semantic_codec = build_semantic_codec(cfg.model.semantic_codec, device)  
    # 3. build acoustic codec  
    codec_encoder, codec_decoder = build_acoustic_codec(cfg.model.acoustic_codec, device)  
    # 4. build t2s model  
    t2s_model = build_t2s_model(cfg.model.t2s_model, device)  
    # 5. build s2a model  
    s2a_model_1layer = build_s2a_model(cfg.model.s2a_model.s2a_1layer, device)  
    s2a_model_full =  build_s2a_model(cfg.model.s2a_model.s2a_full, device)  
  
  
    # load semantic codec  
    safetensors.torch.load_model(semantic_codec, semantic_code_ckpt)  
    # load acoustic codec  
    safetensors.torch.load_model(codec_encoder, codec_encoder_ckpt)  
    safetensors.torch.load_model(codec_decoder, codec_decoder_ckpt)  
    # load t2s model  
    safetensors.torch.load_model(t2s_model, t2s_model_ckpt)  
    # load s2a model  
    safetensors.torch.load_model(s2a_model_1layer, s2a_1layer_ckpt)  
    safetensors.torch.load_model(s2a_model_full, s2a_full_ckpt)  
  
    # inference  
    prompt_wav_path = args.audio  
    save_path = "output.wav"  
    prompt_text = args.prompt_text  
    target_text = args.text  
    # Specify the target duration (in seconds). If target_len = None, we use a simple rule to predict the target duration.  
    target_len = None  
  
    maskgct_inference_pipeline = MaskGCT_Inference_Pipeline(  
        semantic_model,  
        semantic_codec,  
        codec_encoder,  
        codec_decoder,  
        t2s_model,  
        s2a_model_1layer,  
        s2a_model_full,  
        semantic_mean,  
        semantic_std,  
        device,  
    )  
  
    recovered_audio = maskgct_inference_pipeline.maskgct_inference(  
        prompt_wav_path, prompt_text, target_text,args.language,args.target_language, target_len=target_len  
    )  
    sf.write(save_path, recovered_audio, 24000)

首次推理会在hf_download目录下载10个G的模型。

推理过程中，会占用11G的显存：

如果您的显存低于11G，那么务必打开Nvidia控制面板的系统内存回退策略，通过系统内存来补足显存：

如果愿意，也可以基于gradio写一个简单的webui界面，app.py:

import os  
import gc  
import re  
import gradio as gr  
import numpy as np  
import subprocess  
os.environ['HF_HOME'] = os.path.join(os.path.dirname(__file__), 'hf_download')  
# 设置HF_ENDPOINT环境变量  
os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"  
  
  
  
reference_wavs = ["请选择参考音频或者自己上传"]  
for name in os.listdir("./参考音频/"):  
    reference_wavs.append(name)  
  
def change_choices():  
  
    reference_wavs = ["请选择参考音频或者自己上传"]  
  
    for name in os.listdir("./参考音频/"):  
        reference_wavs.append(name)  
      
    return {"choices":reference_wavs, "__type__": "update"}  
  
  
def change_wav(audio_path):  
  
    text = audio_path.replace(".wav","").replace(".mp3","").replace(".WAV","")  
  
    # text = replace_speaker(text)  
  
    return f"./参考音频/{audio_path}",text  
  
  
def do_cloth(gen_text_input,ref_audio_input,model_choice_text,model_choice_re,ref_text_input):  
  
    cmd = fr'.\py311_cu118\python.exe local_test.py -t "{gen_text_input}" -p "{ref_text_input}" -a "{ref_audio_input}" -l {model_choice_re} -lt {model_choice_text} '  
  
    print(cmd)  
    res = subprocess.Popen(cmd)  
    res.wait()  
      
    return "output.wav"  
  
  
  
with gr.Blocks() as app_demo:  
    gr.Markdown(  
        """  
项目地址:https://github.com/open-mmlab/Amphion/tree/main/models/tts/maskgct  
  
整合包制作:刘悦的技术博客 https://space.bilibili.com/3031494  
"""  
    )  
    gen_text_input = gr.Textbox(label="生成文本", lines=4)  
    model_choice_text = gr.Radio(  
        choices=["zh", "en"], label="生成文本语种", value="zh",interactive=True)  
    wavs_dropdown = gr.Dropdown(label="参考音频列表",choices=reference_wavs,value="选择参考音频或者自己上传",interactive=True)  
    refresh_button = gr.Button("刷新参考音频")  
    refresh_button.click(fn=change_choices, inputs=[], outputs=[wavs_dropdown])  
    ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")  
    ref_text_input = gr.Textbox(  
        label="Reference Text",  
        info="Leave blank to automatically transcribe the reference audio. If you enter text it will override automatic transcription.",  
        lines=2,  
    )  
    model_choice_re = gr.Radio(  
        choices=["zh", "en"], label="参考音频语种", value="zh",interactive=True  
    )  
    wavs_dropdown.change(change_wav,[wavs_dropdown],[ref_audio_input,ref_text_input])  
    generate_btn = gr.Button("Synthesize", variant="primary")  
  
      
    audio_output = gr.Audio(label="Synthesized Audio")  
  
    generate_btn.click(do_cloth,[gen_text_input,ref_audio_input,model_choice_text,model_choice_re,ref_text_input],[audio_output])  
      
def main():  
    global app_demo  
    print(f"Starting app...")  
    app_demo.launch(inbrowser=True)  
  
  
if __name__ == "__main__":  
    main()

当然，别忘了安装gradio依赖:

pip3 install -U gradio

运行效果是这样的：

结语

MaskGCT模型的优势在于语气韵律层面十分突出，可以媲美真实语音，缺点也很明显，运行成本偏高，工程化层面优化不足。MaskGCT项目主页中已经有其商业版本模型的入口，据此推断，官方应该不会在开源版本中太过发力，最后奉上一键整合包，与众乡亲同飨:

MaskGCT一键包整合包 https://pan.quark.cn/s/e74726b84c78