Automatically transcribe audio from YouTube videos without captions

httpjames

Many YouTube videos don't have captions, auto-generated or manually transcribed, which prevents Universal Summarizer and Discuss Document from accessing the contents of the video.

Example:

Both AI products already have a feature that transcribes audio files to feed into the LLMs, so it should be applied to YouTube videos as well. Audio tracks from videos can be extracted using something like youtube-dlp.

Vlad

httpjames Transcription is very expensive about 30 cents per average video. We are already absorbing this cost for podcasts, but they are summarized much less frequently than YouTube videos. Currently it is a bit scary (financially) to transcribe youtube videos sent to summarizer.

ForumNinja404

Perhaps this could be achieved by using the open subtitles API if transcribing audio is computationally expensive. I’m not sure what the pricing is, and certain videos may need a fallback if no subtitles are available.

Thibaultmol

Is there a central place for (proper Whisper, not google's auto transcribe stuff which is still not as good somehow?) subtitles for youtube videos? opensubtitles claims to support youtube videos in their database (not referring to their service, just the actual database). But I can't seem to find any videos on there nor how to add subtitles for a youtube video.

It's very computationally intensive obv but I wouldn't mind using my pc in idle to run whisper (like how foldingathome works) but if there's no central place to store them all, then yeah....

Thibaultmol

Also useful to know: https://filmot.com/
this exists. It seems like someone already made a search engine for youtube subtitles. So maybe there's a collaboration possible there