Identifying non-speech?

[icsy7867]

So this is something I am trying to do to learn and to build a dataset.

This is a bit of a ridiculous thing I'm trying to do, but it's been very educational.

Long story short, let's say I have a ton of audio clips of people speaking and animal noises. In this particular case, people are talking, but if there are lions roaring between the people talking I want to insert (test).

(I'm experimenting with creating my own special token and training a tts model).

I have a feeling that something like this would require fine tuning, but just thought I'd ask!

[icsy7867]

Actually thinking about this....

Maybe it would be better to just take the lion audio, split it into well defined sections, and then combine it with people talking so that I know programmatically the transcription of the audio.

[whicks1]

Try the whisper-at variant. YMMV.

[icsy7867]

Thanks!