Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I really appreciate the effort that went into the performance part of this work. There was a real effort to try and make it sound like a reasonable representation of humans playing...a little off beat, out of sync at times. Instead of just hammering the notes out like I hear with lots of these systems, it makes it listenable...I've had the endless trad on for 15 minutes now in the background.

I also like how the basic structure of the musical forms has mostly carried through the model, that seems to be a good "sniff test" if the model is producing reasonable output, if the musical structure makes sense as well as the notes. It makes it feel like there was a little bit of planning.

Great work.



>I really appreciate the effort that went into the performance part of this work. There was a real effort to try and make it sound like a reasonable representation of humans playing...a little off beat, out of sync at times.

That said, there's a certain ways that humans interact when they play together (even if they track their parts independently).

Just being a little off beat randomnly doesn't capture that, and can sound fake just as being perfectly on beat.

There are a few algorithms about how actual players interact, here's a relevant study: http://scitation.aip.org/content/aip/magazine/physicstoday/a...

There are several more for real-life like quantization/humanization.




Consider applying for YC's Summer 2026 batch! Applications are open till May 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: