The same could be applied to their music engine. If they wanted, they could have hired musicians instead of scrapping streaming songs to train it. But we all know how it started and where it ended.

 

This is exactly how eleven labs does it. Hires musicians to make their music model. It also sounds pretty unique due to it having custom data. Its still a lil sterile but getting there.

 

Oh yeah! i have made custom fine tunes with it.. It actually improves the model. At a base model it sounds like suno 3.5 era.. but with a fine tune on a custom data set i give it of good quality audio. it sounds closer to current suno. Less smearing of drums and clean output!

Nova said they will be doing fine tunes soon too. so might just be easier to do in there.

 

fully agree every company scrapes user data. Malice can be part of it, but also product improvement can as well. Im part of the ableton beta program and they told me years ago that they figured out how to get the click rate down on certain tasks due to the data collected.

 

Well, there's no clear equivalent of male waifu in Japanese, so english-speaking people made that up.

A funny thing is in Japan they consider cool to throw English words in a sentence, which is cringe. That's called Engrish.
And like you mentioned in Japanese certain phonemes/orthography rules are so hard they use it with English even knowing it's wrong. That's why they called it "husband-o"
Happens specially with words ending in "t". If you ask for "toilet?" in a Japanese restaurant nobody would understand. But if you ask "toilet-o?" many will.

Ah, Japan... wait, what was the topic again?