Hacker News Reader

Comment by jefftk | Hacker News Reader

I don't know the other genres well enough to evaluate them, but the bluegrass one (https://soundcloud.com/openai_audio/genre-bluegrass) is pretty bizarre:

* Who uses piano as the lead instrument in bluegrass?

* They're only using one one note velocity for the whole piece, which misses a huge wealth of variation through rhythmic accent.

* Timing generally feels a bit robotic?

* The best parts it sounds kind of ok but boring; in bad parts it sounds like nothing (https://soundcloud.com/openai_audio/gaga-beatles is even worse that way)

* A lot of the artificiality seems to be the synth they're using.

* On the other hand it does understand a bit about phrasing and repetition, which many people new to traditional music take a long time to pick up.

View on HN

It's midi. It's meant to be notation, not a realistic composition. I suppose the background "machine gun" piano notes would be a bassy synth combined with 90s electronic music drums. So please do not focus that much on the audio or performance part of it, but on the composition part of it.

I do agree they could have trained it about the importance of velocity though. (That neural net, and most young music students out there too, heh)

>It's midi. It's meant to be notation, not a realistic composition.

MIDI certainly can represent realistic compositions, when used with good synths, though I agree it's not what it's known for!

> I suppose the background "machine gun" piano notes would be a bassy synth combined with 90s electronic music drums.

Uh, what sort of bluegrass music do you listen to and where can I find some of it?

I think they meant the gaga-beatles one, there's a lot of "machine gun" piano sounds.

https://soundcloud.com/openai_audio/gaga-beatles

Do you think there was enough velocity info in the training corpus?

Interesting question. I would assume yes. There should be enough data for things like sustain pedal and velocity dynamics. Music people can be "crazy" and add a lot of detail in their work. It is pretty common to find midi renditions of pieces that contain all kinds of extra information, from pitch shift cues for guitar solos to legato for string instruments (when the finger slides across the string, instead of "jumping" from position to position), to even volume automation for fade in/out effects. Classical musical notation also contains it, though way more limited than the 0-127 range of MIDI, in the piano/forte dynamics notation.

It might have been stripped off to ease complexity, or maybe it was unadequate and created odd sounding results. Maybe the midi library they used doesn't do a good job of portraying it and it is actually there! Who knows!

The opening sounds a lot like Fisher's Hornpipe [1]. And then at 0:26 and even more so at 0:40 it sounds less and less like bluegrass. The weird instrumentation makes it especially strange.

[1]: https://open.spotify.com/track/4QGO8m9OCsiW5iNF4ZNqgk?si=E8C...

Interesting. I’m not where I can test this out and my first thought was wondering how well MuseNet could do bluegrass. One of my favorite things to look for on YouTube are bluegrass covers of popular songs.

Another interesting musical form to see how well MuseNet does would be boogie woogie.

Reply on news.ycombinator.com