If we’ve learned one thing about artificial intelligence, machine learning, and music generation, it’s that AI makes some damned fine death metal.

I mean, sure, part of why machine learning doesn’t really replace humans in its present form is this very phenomenon you’re hearing. If you put audio content of pre-existing music into a blender and then mathematically spew bits of it out, you’re losing all the nuance of form and compositional intent that make a lot of music genres work in the first place.

But back up – what was that bit about spewing things out of a blender?

If that part sounded like a feature, not a bug, then you sound like just the sort of person who will love AI-generated death metal. I am that sort of person, despite being about nuance and form and compositional intent, uh, most of the time. (At least I normally pretend.) And I’ve written about this before.

But now, it’s worth mentioning because HUMAN EXTINCTION PARTY ๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค–๐Ÿ”ช๐Ÿ˜ต๐Ÿ—ก๏ธ๐Ÿค– is amazing.

It’s a livestream of this stuff, just when the creators, Dadabots, have taken down a lot of their other streams. I realize this came out on the 25th of October, ostensibly for Halloween, which now is of course past. But I would implore Databots to keep this up, and the AI-generated text playing karaoke-style over images of meat, on through Christmas. (Honestly, this makes me feel a lot less murderous than hearing “Last Christmas.” If I seem to be getting stabby and Wham! is on, play this to calm me down and watch the butcher knife slide from my placid fingers.)

And yes, you should read their research paper:

Generating Albums with SampleRNN to Imitate Metal, Rock, and Punk Bands

More on SampleRNN:

http://deepsound.io/samplernn_first.html

I still think this will not really turn into a generative model for music, but could turn into a far more interesting way of processing your own samples than only looping them, by generating larger-scale textures out of existing material. If I’m wrong, you can flay my skin and … okay, now I’ve been listening to too much of this stream.

I’m in the midst of our AI Music Lab with MUTEK.jp, so more on this topic – and not just skin deep – soon.

Go go Databots:

http://dadabots.com/faq.php

And yes, if you really want to have an argument about authorship and this stuff, you should probably go talk to your MPC, too.

Also, Hatebeak forever. (And yeah, CDM has been going for 15 years – and if you got the Hatebeak reference, probably you’ve been reading roughly that long.)