from Hacker News

An AI version of Al Michaels will deliver Olympic recaps on Peacock

by prossercj on 6/26/24, 7:41 PM with 2 comments

• by mikemitchelldev on 6/26/24, 7:48 PM

  This is cool from the tech vantage point. But the labor-sympathizer in me would like to see a person employed to do it rather than licensing the right to their image.

• by ilaksh on 6/26/24, 8:38 PM

  They make it sound like they invented their own TTS model or something. I wonder, did they actually, or is this Eleven Labs, some other API, or Style TTS2 or something?

  I mean, it seems like there are a ton of papers for attempts at realistic TTS, but hard to find something really equivalent to the Eleven Labs voice clone that doesn't have a non-commercial restriction on the weights or code. Maybe they really did train a model from scratch?

  It is a really big company. Maybe they have resources for that.