How do I disable non-speaking noise descriptions?

[gth001]

I only want speaking to become text, not (buzzing) (clicking) etc...

[indokid9999]

Try using sed .
For descriptive sounds that have square brackets try

sed "s/.[.*]/ /" <yourinputfile.

for desciptive sounds that have normal brackets try

sed "s/.(.*)/ /" <yourinputfile.

For files that have both types of brackets, try

sed -e "s/.(.)/ /" -e "s/.[.]/ /" <yourinputfile.

[dgoryeo]

I haven't worked with diarization. I tried out few implementations that use pyannote.audio but somehow it doesn't work very well for me --I primarily work with Japanese audio.

[aehlke]

May I ask what you're working on? I'm also dealing with Japanese audio, https://reader.manabi.io

[dgoryeo]

Games and anime. In those content usually the back ground noise, sounds and music make whisper hallucinate. My current flow is to run 2 transcriptions, one with high vad threashold of 8, and the other with low threshold of 2, then to merge the results. Would love to hear about your experience, tips and tricks.

[dgoryeo]

PS. @aehlke , just checked out the reader.manabi.io, nice.

[aehlke]

I've shared my source here https://github.com/lake-of-fire/SwiftWhisperStream which is for Swift and in a state of flux. Nothing unique yet besides the swift streaming implementation and ios + mac targets, and libfvad which I need to replace.

I'm going to add youtube and mpv features to manabi reader so I want to figure out how to run their audio through whisper to get better transcripts if possible