SUNO is killer!

[ghettosynth]

I am not the same, I am now certified 20% robot, minimum. It might be more. Brought to you by Carl's Junior.

[el-bo (formerly ebow)]

Like what? He is not the same?

Hell no!!!

Once renowned for having such exacting standards when it came to filter topologies, he claimed he’d walk off the dance floor if he heard a poor filter sweep. Now we’re supposed to believe the same ghettosynth is listening to lowest common denominator lo-fi hip hop slop.

These ARE NOT the same people!!

[Innermost]

I am not the same, I am now certified 20% robot, minimum. It might be more. Brought to you by Carl's Junior.

ghettoGPT

[ghettosynth]

Like what? He is not the same?

Hell no!!!

Once renowned for having such exacting standards when it came to filter topologies, he claimed he’d walk off the dance floor if he heard a poor filter sweep. Now we’re supposed to believe the same ghettosynth is listening to lowest common denominator lo-fi hip hop slop.

These ARE NOT the same people!!

I think that you are mixing incompatible contexts. I very much care about the sound quality when I'm on the dance floor. The music is loud, often played through unforgiving horn tweeters, and some of the artifacts of things that bug me explicitly are amplified there. At the same time, that is not the state of affairs when I'm trying to get work done and have music playing that sits just below the level of conscious awareness.

For the former, I want to feel every beat and I want the characteristic cream of a filter matched to the application. For the latter I just don't want the tracks to disturb me. Just enough interest so that I don't turn it off, but not so much that it causes me to engage the music for more than a split second or two at a time.

Years ago I worked at home for a year and this was long before streaming. So, to easily listen to music all day the radio was the only real choice. So, I chose the local public radio station that played jazz. The thing is, at that time, I didn't really recognize a lot of the standards by name. By the end of the year though, I did, and they became distracting. I don't want to know the artists or the songs, and I don't want to memorize them over time while I'm working. I think that this is the kind of thing that AI is going to foster at low cost.

As far as underground music, I'm not convinced that any of the models are trained on enough to do a good job TBH. They do ok at the harder styles, but they lack the subtlety for the kind of house/techno that interest me. Current research identifies this as a problem. It requires the model to represent a persistent source plus explicit, temporally varying control states. Systems such as JASCO introduce fine-grained temporal conditioning precisely because global text descriptions are too blunt for this sort of control. Related, it's just too much to infer about the underlying model about the characteristics of the filter sweep. So, they tend to sound tame and boring. Not distorted, per se, just not interesting.

So, no, I am not convinced by SUNO 90s house, it's largely terrible and either comes out chill, or way too modern sounding. Now, where it matters much less is in muted mostly convincing acoustic sounds, e.g., LoFi hip hop, for when I'm studying. It works reasonably well for those and adjacent genres.