News for hackers. Stuff that matters.
Re:AI Is Destroying the Trust Between Writers and Readers by zahlman on Friday November 13, 2026 @11:28AM (#50485863)
> but also the actual sound quality on a lot of AI generated music is awful.

I feel like I don't quite know what to listen for here.

Re:AI Is Destroying the Trust Between Writers and Readers by komali2 on Friday November 13, 2026 @11:40AM (#50486046)
Hi-hats. When testing sound quality, pay attention to the hi-hats. Once you notice the shimmer introduced by audio compression to a hi-hat you can't unhear it, and for whatever reason AI generated music replicates this.
Re:AI Is Destroying the Trust Between Writers and Readers by TheCapn on Friday November 13, 2026 @12:59PM (#50487242)
>and for whatever reason AI generated music replicates this.

Its within the training set I believe. The AI is being trained off compressed audio

I was listening to a podcast last year (forget which, or I'd attribute credit) and one of the tells AI researchers look for when trying to determine if music is AI generated or not is the compression artifacts. AI likely has gotten better at hiding it since then, but AI would include compression artifacts on the final product because of the prevalence of it in training data.

Re:AI Is Destroying the Trust Between Writers and Readers by zahlman on Saturday November 14, 2026 @11:14PM (#50503959)
I agree with the other poster that this must be because the training data is full of the specific compression artifact you have in mind.

The problem is, I just... don't hear those things. I used to listen to 128kbps MP3s all the time and legitimately not understand wtf my audiophile friends were on about when they would say the quality was terrible. I come from a very musical background and am at least quite close to "perfect pitch", but apparently my ability to interpret sound as music has come at the expense of recognizing the sound for what it really is. :/

Re:AI Is Destroying the Trust Between Writers and Readers by komali2 on Sunday November 15, 2026 @05:06AM (#50505609)
I mean if you don't care you don't care, which is great because you'll save a lot of money :p

I think some people just can't unhear the compression artifacts. The equipment matters a lot. Bluetooth headphones from a phone, mp3 vs flac pretty much indistinguishable. Same as those little ipod headphones.

Once you get to DAC and IEM territory, the difference is quite a lot more noticeable. So for a lot of us it's not just file format but also the equipment to take advantage of it.

Re:AI Is Destroying the Trust Between Writers and Readers by californical on Friday November 13, 2026 @11:42AM (#50486062)
There are a couple sound-quality based tells that I've noticed, off the top of my head

1. Usually weird little "flickers" of sound - like how a candle gives you mostly constant light but there are little flashes of bright and dark. AI generated music tends to have little distortions scattered throughout like that. As the models have improved they're less obvious but still absolutely there if you try to listen to any individual 'instrument' or 'track'

2. very low dynamic range - loud sounds and quiet sounds are much more similar volume than you'd expect from most real music, and things that are supposed to fade or taper-off in volume are unnatural

Re:AI Is Destroying the Trust Between Writers and Readers by zahlman on Saturday November 14, 2026 @04:24AM (#50495463)
> very low dynamic range

... Isn't that just how music is (increasingly) produced nowadays? Did the Loudness War ever actually end?