[FEATURE REQUEST] Audio Splitting more accurately with an llm

[GoudaCouda]

Would it be possible to implement an llm(even a very small one) to help split the audio files. I think with the transcriptions maybe we can feed it a prompt to determine normal stopping, breath, and change of sentences to aid in the splitting of audio samples. I think implementing a weighting system with this and the standard process could make for some more efficient samples.

[erew123]

Audio splitting in what? Finetuning? If so, someone else is doing work on that at the moment #419

[GoudaCouda]

Yea on the fine tuning

[erew123]

Ok, well it currently uses Whisper, which is a LLM in effect https://github.com/openai/whisper?tab=readme-ov-file#whisper and also Silero https://github.com/snakers4/silero-vad?tab=readme-ov-file#silero-vad and someone else is looking at other ways of splitting. I hope that answers the question?

Thanks