Hacker News new | ask | show | jobs
by cptskippy 1340 days ago
> Couldn't this guy just hire a professional captioning service or offer transcripts to users that may need it?

Yes, and that is neither cheap or easy.

2 comments

It is actually quite inexpensive for most and very easy, but expense is of course always measured in relative terms.
Is it easy though? I had a professor do it for his own videos because ‘it is easy’; it is a horror show; everything is badly out of sync etc. I don’t think it’s easy for an average person at all. I am a programmer but I have 2 left hands where it comes to video or photo editing and then it gets expensive fast. People here saying ‘it costs only $150/hr to fix’ probably don’t provide anything free nor have a small business. Throwing 1000s against something that will never make you a penny but is a hobby/something you like because of some lawyers is insanity.

I for instance script my videos, so I have the subtitles already, I just have no way of putting them in so I provide the transcript. Which seems fine for people.

I can't speak to your professor, but we regularly have many dozens of films subtitled and so far we've had sterling results and no complaints from our audience.

Agreed with the cost impact on a small business or nonprofit. We make money on the films so it's a no-brainer for us (and is legally required in places like Canada).

Whisper is free and classroom instruction should be a breeze and require little correction, as there likely isn't much background noise or music.
And it only works with English. As discussed elsewhere in the thread, the majority of the videos are in Hebrew.

Even for much less exotic languages such as French automatic speech recognition systems don't work or are extremely bad.

I mentioned Whisper because it works with a lot of languages. But I understand your confusion, because there are additional lightweight models that are only available for English. Its accuracy is less good for Hebrew, but instructional materials are likely optimal input.

https://github.com/openai/whisper

> And it only works with English.

Depending on how you define "work", Whisper also works with Hebrew. Not sure if the word error rate is acceptable though https://github.com/openai/whisper/#available-models-and-lang...