v3.99

twentyhq/twentyv3.99Apr 14, 2026by jianchang512

AI Summary

This update focuses on audio processing improvements by introducing new AI models for noise reduction and voice separation, optimizing caching strategies, and unifying the user interface for reference audio settings.

Key Highlights

New noise reduction model (dpdfnet) and voice separation model (spleeter)
Optimized caching strategy to maximize reuse of dubbing cache
Unified reference audio settings interface across TTS engines
Added debug.exe for troubleshooting unclear error messages

New Features

dpdfnet noise reduction model support
Spleeter voice separation model support
Unified reference audio settings UI
Debug.exe utility file
Optimized default settings and removal of unsupported AI models
Updated subtitle translation prompt with terminology table variable

Full Release Notes

*v3.99 0508更新日志*

- 降噪模型改用 [dpdfnet](https://k2-fsa.github.io/sherpa/onnx/speech-enhancement/dpdfnet.html)
- 人声背景声分离模型增加 [spleeter](https://k2-fsa.github.io/sherpa/onnx/source-separation/models.html#spleeter)
- 优化缓存策略，以便尽量多使用配音缓存(未关闭软件并对同一文件反复处理时)
- 统一参考音频设置界面(F5-TTS/CosyVoice/Chatterbox/OmniVoice/Index-TTS/VoxCPM/Moss-nano)
- 增加 _debug.exe 调试文件，当出错但报错信息不明时，可使用该文件启动
- 优化一些默认设置项，移除已不再支持的某些AI模型名字
- 修改字幕翻译提示词中`术语表`嵌入变量





##  v3.99 完整包下载(2.5G)


百度网盘下载地址: https://pan.baidu.com/s/1HN3qVb5MMrEf8ukce6D-tA?pwd=1234

huggingface.co:  https://huggingface.co/mortimerme/repocollect/resolve/main/win-pyvideotrans-v3.99-0508.7z?download=true


<details>
<summary>v3.99 补丁包(300MB)</summary>

> 不可单独使用，仅用于覆盖旧包，与完整包区别是不含有`torch`模块
> 因torch从 2.5 升级到 2.7，补丁包不含有torch，直接下载覆盖补丁包可能导致无法cuda加速，建议下载完整包覆盖
> **覆盖后如果闪退，请下载完整包**

百度网盘下载地址: https://pan.baidu.com/s/1iLcbVLt_cSOYaMlCvgj1rw?pwd=1234

huggingface.co: https://huggingface.co/mortimerme/repocollect/resolve/main/win-PatchUpdate-v3.99-0508.7z?download=true

</details>