search

chidiwilliams / buzz

Buzz transcribes and translates audio offline on your personal computer. Powered by OpenAI's Whisper.
https://chidiwilliams.github.io/buzz
MIT License
12.41k stars 938 forks source link

0.9.0 Long audio to subtitle, with advertisements #726

Closed XLiming closed 6 months ago

XLiming commented 6 months ago

When converting audio files to subtitles, they may become advertisements after 16 minutes, prompting likes and follow. What should be done?

chidiwilliams commented 6 months ago

Yes, this is a hallucination from the Whisper model itself, not Buzz. You can learn more about it here and here. You can try some things, like using a different model (larger/smaller), splitting the audio into smaller chunks, etc., but it's a known problem with these models.

XLiming commented 6 months ago

Thank you very much!