Transcription contains duplicated fragments about whisper-timestamped

[Jeronymous]

This is probably due to the fact that whisper use temperature fallback and beam search, while it's currently not supported by whisper-timestamped (see issue #10, it's WIP).

Can you please try whisper with options

--beam_size None --temperature_increment None

and tell if you see the duplicate fragments?

If you see it, then your problem will be solved when issue #10 will be solved

[Jeronymous]

@romanzoniit You can now use option --accurate in whisper_timestamped to reproduce default options of whisper (see README for more information).
Can you please check if that solves this issue for you?

[Jeronymous]

I think this issue is solved. Don't hesitate to re-open if it's still an issue after my last comments

[romanzoniit]

@romanzoniit You can now use option --accurate in whisper_timestamped to reproduce default options of whisper (see README for more information). Can you please check if that solves this issue for you?

how to use this in python, not in console

[Jeronymous]

It's explained in the README

By default, all options that require several steps of decoding are disabled, in favour of an efficient decoding strategy. Use beam_size=5, best_of=5, temperature=(0.0, 0.2, 0.4, 0.6, 0.8, 1.0) for Whisper default.