ReaLLMASIC / nanoGPT

The simplest, fastest repository for training/finetuning medium-sized GPTs.
MIT License
23 stars 17 forks source link

Add support for Audio Tokenization and Voice Emulation via SNAC Tokenization #198

Closed klei22 closed 1 month ago

klei22 commented 1 month ago

This PR creates a new folder in the data directory for snac processing.

See README.md for 3 step demo how to emulate voice timbre based on a collection of mp3 files.

Current settings are created for voice, but we can combine this with other methods for further reducing context length.