Save fully processed chunks to optimize decoding

xenova / whisper-web

ML-powered speech recognition directly in your browser

https://hf.co/spaces/Xenova/whisper-web

MIT License

1.29k stars 152 forks source link

Save fully processed chunks to optimize decoding #28

Closed shopped closed 4 months ago

shopped commented 4 months ago

Saw some TODO comments while looking through the repo, thought I would address them The performance boost is minor, and only when processing long files

I'll add comments to my code changes

I welcome any and all feedback!

shopped commented 4 months ago

Dug deeper into the code and found another issue upstream, will potentially re-open this one later