G1017

请问您解决了吗？是因为录音时长太长了嘛？

/CosyVoice/cosyvoice/cli/cosyvoice.py ```python # Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu) # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except...

增加/CosyVoice/cosyvoice/llm/llm_vllm.py ```python import time import queue import asyncio import threading from typing import List, Generator, AsyncGenerator import torch from cosyvoice.utils.file_utils import logging from cosyvoice.llm.llm import Qwen2LM # # 启用vllm V1版本...

> > > > 看看[https://github.com/qi-hua/async_cosyvoice的项目，VLLM能跑通](https://github.com/qi-hua/async_cosyvoice%E7%9A%84%E9%A1%B9%E7%9B%AE%EF%BC%8CVLLM%E8%83%BD%E8%B7%91%E9%80%9A) > > > > > > > > > 这个试过啦，已经跑不通更新的模型了 > > > > > > model最新的0.5B可以用。但是那个项目是复制了一份推理代码，没有用最新基座的推理代码，反正能跑。等官方合并 > > 问下你跑出来的推理结果是正常的吗，上面这个项目我适配了一下最新的推理代码0.5b推理出来的结果不正常，音频中噪声非常大。 我也是定位到token生成错误

> Could you please translate this issue to English? Hello, it has been changed to English

你可以参考我的文章https://blog.csdn.net/qq_45080326/article/details/147513802?spm=1001.2014.3001.5501

复制300m-sft的这个文件可以用

**搭配使用仍然报错** 以下为示例： **示例1：** File "/root/apps/metahuman/creator/test.py", line 54, in use_historical_model tts_historical( File "/usr/local/lib/python3.9/site-packages/paddlespeech/cli/utils.py", line 328, in _warpper return executor_func(self, *args, **kwargs) File "/usr/local/lib/python3.9/site-packages/paddlespeech/cli/tts/infer.py", line 710, in __call__ self.infer(text=text, lang=lang, am=am, spk_id=spk_id)...

使用的是paddlespeech-r1.4版本

> 目前报错为xml解析错误，您可能需要规范一下输入的文本格式。需要搭配出现。 麻烦您看一下上面的回复，谢谢~