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

Some of word not splitting #657

Closed prithiviclteam closed 2 months ago

prithiviclteam commented 10 months ago

Hi,

I have tested your Buzz app, and the data quality is good. However, some lines are not splitting in the output. For example below and I have attached respative files and mp3 files for testing:

22 00:00:11,940 --> 00:00:13,169 of Mum's flowers,

27 00:00:10,679 --> 00:00:10,769 many

28 00:00:10,769 --> 00:00:11,360 kids with one leg kicking footballs he thought.

[Buzz_Testing.zip](https://github.com/chidiwilliams/buzz/files/13792157/Buzz_Testing.zip)

raivisdejus commented 2 months ago

How words and sentences are returned depends on the model configuration during training. If some model does not split words or sentences as you need you can try different model. Discussion thread to collect good custom models is here https://github.com/chidiwilliams/buzz/discussions/866

Also in the upcoming 1.0.2 version there will be a feature that lets you resize subtitles after they have been generated. This feature is available in latest development version. See artifact section of the development builds https://github.com/chidiwilliams/buzz/actions/workflows/ci.yml?query=branch%3Amain To download log into the github.