Posted
It's already a bit older (2019), but this AI generated (A google /magenta project: Music Transformer)
is the the most authentic music composed by AI (inspired by 10000 hours of human composed music) that me (and probably the world) has ever heard:
Here you can listen to 100.000 files, it's a online player, they claim it's "no cherry picking", just as the AI delivered)
(https://magenta.github.io/listen-to-transformer/#a1_68611.mid)
You can download the file with "share" - >dowload song
Mo., 09.01.2023 - 11:53 Permalink
Thanks for the link. The Magenta project has been around for a while already, but this jukebox is new to me.
Generative A.I. can already do a lot with images (MidJourney) and language (ChatGPT). Music though is a challenge for Deep Learning. Symbolic A.I. is still more capable in discrete temporal domains like music.
If you want to use Magenta material with Synfire, you can also download the original recordings (MAESTRO Dataset), quantize them and run any spans of interest through figure recognition. Then use the Morphing parameter in an arrangement to generate blends of multiple styles.
They also made a few plugins that are fun to play with. Although I think Factories are better suited to do the job. I would not rule out though, that some day we might have Factories that use deep learning models under the hood.
Fr., 22.12.2023 - 11:32 Permalink
f you want to use Magenta material with Synfire, you can also download the original recordings, quantize them and run any spans of interest through figure recognition. Then use the Morphing parameter in an arrangement to generate blends of multiple styles.
Does Synfire offer the ability to do the needed (?) quantization, or do we have to do that elsewhere (e.g. in a DAW)?
Also, what settings would you suggest for the quantization to get good results?