Starstruk Logo MarkStarstruk

v2.0

Fictionarry/TalkingGaussianv2.0Mar 31, 2026by lipku

AI Summary

Major architectural refactoring separating code into specific directories and adopting a plugin-based architecture for models and TTS.

Key Highlights

  • Refactored code structure with models and audio features moved to avatars directory
  • Switched models, TTS, and transport methods to plugin architecture
  • Added RTMP output in streamout directory
  • Integrated Aliyun QwenTTS
  • Unified inference and reply code in BaseAvatar class

New Features

  • Plugin architecture for models and TTS
  • RTMP output support
  • Aliyun QwenTTS integration

Full Release Notes

1，重构代码，数字人模型和音频特征代码移到avatars目录下
2，数字人模型、tts、传输方式改成plugin方式接入
3，传输方式整理成单独的类，放到streamout目录下，添加rtmp输出
4，数字人推理和回贴公共部分代码整合到BaseAvatar中，各模型只需要实现自己特有部分代码
5，音频特征的切分代码整合到BaseAsr中
6，tts添加阿里云qwentts