Wouldn’t building anything with this scrape be a YouTube TOS violation?

You raise a good point. Maybe there should be an effort by creators to publish their videos on other platforms with different licenses too, therefore removing any advantage Google has in training AI using others’ content. Legalities aside, I find it really distasteful that Google can claim exclusivity over a use of the content that they were never owed. Let’s be honest - they probably granted themselves that legal “right” with some obscure update to their terms of service.

Creators are allowed to publish their videos under CC-BY licence.

According to the preview, some of these (30 thousand results?) are "Wikipedia article audios"[1], so text turned into speech and now turned back to text

[1]: https://huggingface.co/datasets/PleIAs/YouTube-Commons/viewe...

I hope folks are not trying to build a "very rational and factual bot" model based on youtube transcriptions.

Indeed it could be useful for "my bot runs its own clickbait channel" type of thing.

If anyone can help me here.

What are cost-effective ways to convert audio to transcript?

It should support some non-english languages such as French, German, Portuguese, Spanish.

OpenAI's Whisper is a good transcribing tool. You can use MacWhisper for free and it uses the OpenAI Whisper technology locally on your Mac.That is if you're using a Mac. I'm not sure about Windows or Linux. https://goodsnooze.gumroad.com/l/macwhisper

Perfect. Thank you!

+1 for running Whisper locally, but check out Assembly AI too: https://www.assemblyai.com/pricing