ChatTTS完美部署

一、项目工程下载

2noise/ChatTTS: ChatTTS is a generative speech model for daily dialogue. (github.com)

直接git clone即可！（直接down包，解压）

二、模型下载

测试了两种，直接编写代码下载即可！最简单方便。

#SDK模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('pzc163/chatTTS')

不写绝对路径就会保存在这个位置

C:\Users\Administrator\.cache\modelscope\hub\pzc163

下载完之后移动到自己的工程下，目录如下：

三、环境安装

需要安装下面的库

omegaconf~=2.3.0
torch~=2.1.0
tqdm
einops
vector_quantize_pytorch
transformers~=4.41.1
vocos

如果是新的环境，就直接全部安装，如果本身有一些库就自行pip install就可以。

全部安装运行：

pip install -r requirements.txt

四、运行Demo.py并保存结果

import scipy

import ChatTTS
from IPython.display import Audio

chat = ChatTTS.Chat()
chat.load_models(source='local', local_path='ChatTTS')

params_infer_code = {'prompt':'[speed_5]', 'temperature':.3}
params_refine_text = {'prompt':'[oral_2][laugh_0][break_6]'}

texts = ["四川美食可多了，[uv_break] 有麻辣火锅、宫保鸡丁、麻婆豆腐、[uv_break] 担担面、回锅肉、夫妻肺片等， [uv_break] 每样都让人垂涎三尺。"]

wav = chat.infer(texts, \
    params_refine_text=params_refine_text, params_infer_code=params_infer_code)

#texts = ["This is a test of the ChatTTS script.  Peter Piper picked a peck of pickled peppers.  Red leather.  Yellow leather.  Red leather.  Yellow leather.  Red leather.  Yellow leather.",]

# wavs = chat.infer(texts, use_decoder=True)
Audio(wav[0], rate=24_000, autoplay=True)
scipy.io.wavfile.write(filename = "output.wav", rate = 24_000, data = wav[0].T)

五、报错情况

windows会报错，需要改ChatTTS/core.py，第75行。

compile: bool = False,

结语：以上内容仅供学习使用！！！