Distinguish actors in the conversation

[atolmachev]

Is there a feature to distinguish actors (person1, person2, ...) so that it will be easier to use the output as the conversation/podcast/coaching session/... transcription?

[elijahfhopp]

Apparently this is called diarizaition. It appears to be available for stereo recordings in the command line:

-di,      --diarize         [false  ] stereo audio diarization

I haven't used it myself, so I can't vouch for its quality.

If you have multi-track audio then you can obviously just use it on each track and then perhaps have a script that takes the timestamps to make them into one, cohesive transcription. I am currently working on a project that will use a system like this, so if you end up writing anything, ping me.

[strangelearning]

Anecdotally, a user and I have tried the same thing and diarization isn't quite working.

#489

We have both converted mono audio files into stereo files using ffmpeg ffmpeg -i input.wav -ac 2 output.wav

And we both get (speaker ?) for our attempts. I'll be tracking the issue linked above.

The commands I am using are ./main -m models/ggml-tiny.bin -f ../../openai/yt-semantic-search/firstep/outputstereo.wav -di and ./main -m models/ggml-tiny.bin -f ../../openai/yt-semantic-search/firstep/outputstereo.wav -di -ml 1 respectively.

[ciaranmays]

is there any way to make it so theres something like Speaker 1, speaker 2, etc?