this post was submitted on 13 May 2025
276 points (94.5% liked)

Technology

70001 readers
4713 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots


founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 13 points 13 hours ago (2 children)

Most of these languages dont even have enough professional voice actors to cover the bandwidth.

And you think anyone is training AI voice models for those languages? Have you even seen how long it takes even large companies like Google to support the languages with hundreds of millions of speakers?

[–] [email protected] 1 points 9 hours ago* (last edited 9 hours ago) (1 children)

That's the benefit of using AI and machine learning - once you have enough source material, you can throw it all in and it'll eventually spit out a model.
Which is exactly what Meta did with their Massively Multilingual Speech project which supports text-to-speech and speech-to-text for 1107 different languages.

Is it actually any good in 99% of them, I don't have a clue, but it exists.

[–] [email protected] 1 points 9 hours ago

Seems more like a proof of concept project for that paper than something they are pursuing seriously judging by the GitHub location in some example folder that hasn't seen any significant updates in over a year. If it is so great I would assume they would pursue it more actively and replace existing models with it two years later.

[–] [email protected] -2 points 12 hours ago (2 children)

It becomes easier and cheaper every day. Today's open source LLMs are better than last year's best model.

[–] [email protected] 4 points 9 hours ago (1 children)

Is it? I just tried again yesterday for a simple script since coding is the one thing apparently AI will replace people like me and it could not put together a working JavaScript script.

I have yet to see tangible results not announced by the people with sunken cost exploding their balls.

[–] [email protected] 0 points 4 hours ago* (last edited 3 hours ago) (1 children)

Sounds like a skill issue my dude. While you struggle to get a js script people are putting out entire programs with AI assistants so sure - you're right and they're wrong

[–] [email protected] 1 points 2 hours ago (1 children)

yeah, I guess I didn't prompt right lol

[–] [email protected] 1 points 1 hour ago

Yes, to effectively use AI you actually have to understand the medium you're in to describe the problem you're trying to solve. You can get there with prompting but it'll take you much longer if you just don't understand code yourself.

Thats why most senior software devs are not afraid of LLMs cause they need strong oversight and thats exactly what years of software dev experience trained you to do.

[–] [email protected] 6 points 11 hours ago (1 children)

You're fundamentally misunderstanding the comment you replied to, they are not saying that voice AI are bad, they are saying there is not enough training data to improve the AI for these languages. How will it improve without good training data?

[–] [email protected] 1 points 4 hours ago

Thats not how AI training works and even then there's absolutely enough data. Also training data can be created and even synthesized. There are many techniques to extract make training value from datasets that we discover every year - It's really not a problem you think it is.

I'm genuinely confused how AI illiterate users here are. It's just blind leading the blind.