前言

传言GPT-SoVITS作为当前与BertVits2.3并列的TTS大模型，于是本地搭了一个，简单说一下坑。

搭建

下载

到GitHub点击此处下载
https://github.com/RVC-Boss/GPT-SoVITS

解压

解压到全英文目录

VSCode打开

使用VSCode打开，切到conda并clone一个之前BertVits的环境（没环境的自己先做一个Python3.10的配好PyTorch的）

安装依赖包

使用下面语句安装依赖

pip install -r requirements.txt

修改内容

根据issues内大家讨论的结果，这样操作是实测可行的，但是之后作者应该会优化，截止发文这么改是没问题的，以后可能不用改了
https://github.com/RVC-Boss/GPT-SoVITS/issues/26

1.重新安装版本

输入下面的指令重新安装一下对应版本的

pip install funasr==0.8.7
pip install modelscope==1.10.0

2.修改文件内容

将 tools\damo_asr\cmd-asr.py 文件中的

inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='tools/damo_asr/models/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
    vad_model='tools/damo_asr/models/speech_fsmn_vad_zh-cn-16k-common-pytorch',
    punc_model='tools/damo_asr/models/punc_ct-transformer_zh-cn-common-vocab272727-pytorch',
)

改成

inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
    vad_model='damo/speech_fsmn_vad_zh-cn-16k-common-pytorch',
    punc_model='damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch',
)

根据我的研究，原因是 git clone 的那几个模型的配置文件和它自动下载的内容不一样

修改后源码如下

# -*- coding:utf-8 -*-

from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
import sys,os,traceback
dir=sys.argv[1]
# opt_name=dir.split("\\")[-1].split("/")[-1]
opt_name=os.path.basename(dir)

path_asr='tools/damo_asr/models/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'
path_vad='tools/damo_asr/models/speech_fsmn_vad_zh-cn-16k-common-pytorch'
path_punc='tools/damo_asr/models/punc_ct-transformer_zh-cn-common-vocab272727-pytorch'
path_asr=path_asr if os.path.exists(path_asr)else "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"
path_vad=path_vad if os.path.exists(path_vad)else "damo/speech_fsmn_vad_zh-cn-16k-common-pytorch"
path_punc=path_punc if os.path.exists(path_punc)else "damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch"
# inference_pipeline = pipeline(
#     task=Tasks.auto_speech_recognition,
#     model=path_asr,
#     vad_model=path_vad,
#     punc_model=path_punc,
# )
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
    vad_model='damo/speech_fsmn_vad_zh-cn-16k-common-pytorch',
    punc_model='damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch',
)

opt=[]
for name in os.listdir(dir):
    try:
        text = inference_pipeline(audio_in="%s/%s"%(dir,name))["text"]
        opt.append("%s/%s|%s|ZH|%s"%(dir,name,opt_name,text))
    except:
        print(traceback.format_exc())

opt_dir="output/asr_opt"
os.makedirs(opt_dir,exist_ok=True)
with open("%s/%s.list"%(opt_dir,opt_name),"w",encoding="utf-8")as f:f.write("\n".join(opt))

运行

在环境中输入，即可正常启动

python webui.py

总结

能够有感情的朗读了，不错