collabora / WhisperLive

A nearly-live implementation of OpenAI's Whisper.
MIT License
2.09k stars 283 forks source link

About the no_speech_prob results returned by the faster_whisper #291

Open giaoyyds opened 3 weeks ago

giaoyyds commented 3 weeks ago

I created a py script that worked for me by following the logic of the server.py in the whisper_live, and there was a problem: when I called the transcriber.py to transcribe, the correct text was returned, but the no_speech_prob always returned 0, causing me to not be able to filter based on no_speech_prob.

Returned results : [Segment(id=1, seek=429, start=0.0, end=3.12, text='お電話ありがとうございます', tokens=[50365, 6117, 20545, 11103, 40035, 50521], temperature=0.0, avg_logprob=-0.3460582277991555, compression_ratio=0.7924528301886793, no_speech_prob=0.0, words=None), Segment(id=2, seek=429, start=3.12, end=4.04, text='FPI', tokens=[50521, 37, 31701, 50567], temperature=0.0, avg_logprob=-0.3460582277991555, compression_ratio=0.7924528301886793, no_speech_prob=0.0, words=None)]