I added new features: mp3-extend and google colab

[Mozer]

Thanks for a great project!

I added new features:

• --extend_mp3 (to continue any uploaded split mp3 with voice cloning)
• --resume_after_n (to continue previous generation)
• created a free colab: https://colab.research.google.com/github/Mozer/YuE-extend/blob/main/colab/Yue_extend_with_exllama.ipynb

It is using:

• exllama module implemented by https://github.com/sgsdxzy/YuE-exllamav2
• web UI by https://github.com/alisson-anjos/YuE-exllamav2-UI

Notes:

• changes are implemented in exllama workflow. Original HF workflow is not implemented (todo).
• --extend_mp3 works best with segments <= 30s. Long mp3s can cause OOM error. I recommend extending right after first verse end. Put needed seconds into Seconds to take from mp3
• --extend_mp3 takes 2 separate tracks as input: vocal.mp3 + instrumental.mp3. To split your mp3 use: python-audio-separator or other services
• extend_mp3 works with COT and ICL models
• seeding is currently not working with exllama

Code: https://github.com/Mozer/YuE-extend

[jrked]

great work!❤️

[a43992899]

Thx! Just added your repo to readme.