[Csnd] Machine learning for composing -- where are we at?

Date	2020-05-08 15:20
From	Michael Gogins
Subject	[Csnd] Machine learning for composing -- where are we at?
	TL;DR: Can generate songs up to minutes in length -- as raw waveforms. With vocals, lyrics, instruments, and all. Audio quality is not up to snuff. But still, VERY VERY IMPRESSIVE. This, insofar as my limited expertise in this domain is concerned, is one of several projects (like Magenta) that represents the current state of the art in the use of artificial intelligence/machine learning methods to generate music from scratch. A snapshot (I would guess out of date) of the code is at https://github.com/openai/jukebox. The repository has a link to a paper which explains how it works, to sample pieces, and to a colab notebook. This is a company, a very interesting company (https://openai.com/about/). And the software license reflects that. I, as a composer and programmer, could freely use this software, modify this software, and publish music made using this software -- BUT NOT IF I EVER WANT TO MAKE ONE RED CENT OFF MY MUSIC. Presumably I could pay for a license, but there's no obvious way to do that. Maybe that's just for the investors in this company. That's too bad, but this is still very interesting, and it should be possible (but maybe not easy) to create similar software based on the ideas in this software that would be truly open source. The lead author is clearly a live wire. Regards, Mike ----------------------------------------------------- Michael Gogins Irreducible Productions http://michaelgogins.tumblr.com Michael dot Gogins at gmail dot com Csound mailing list Csound@listserv.heanet.ie https://listserv.heanet.ie/cgi-bin/wa?A0=CSOUND Send bugs reports to https://github.com/csound/csound/issues Discussions of bugs and features can be posted here

Date	2020-05-08 16:44
From	Richard
Subject	Re: [Csnd] Machine learning for composing -- where are we at?
	This is amazing! On 08/05/2020 16:20, Michael Gogins wrote: TL;DR: Can generate songs up to minutes in length -- as raw waveforms. With vocals, lyrics, instruments, and all. Audio quality is not up to snuff. But still, VERY VERY IMPRESSIVE. This, insofar as my limited expertise in this domain is concerned, is one of several projects (like Magenta) that represents the current state of the art in the use of artificial intelligence/machine learning methods to generate music from scratch. A snapshot (I would guess out of date) of the code is at https://github.com/openai/jukebox. The repository has a link to a paper which explains how it works, to sample pieces, and to a colab notebook. This is a company, a very interesting company (https://openai.com/about/). And the software license reflects that. I, as a composer and programmer, could freely use this software, modify this software, and publish music made using this software -- BUT NOT IF I EVER WANT TO MAKE ONE RED CENT OFF MY MUSIC. Presumably I could pay for a license, but there's no obvious way to do that. Maybe that's just for the investors in this company. That's too bad, but this is still very interesting, and it should be possible (but maybe not easy) to create similar software based on the ideas in this software that would be truly open source. The lead author is clearly a live wire. Regards, Mike ----------------------------------------------------- Michael Gogins Irreducible Productions http://michaelgogins.tumblr.com Michael dot Gogins at gmail dot com Csound mailing list Csound@listserv.heanet.ie https://listserv.heanet.ie/cgi-bin/wa?A0=CSOUND Send bugs reports to https://github.com/csound/csound/issues Discussions of bugs and features can be posted here

Date	2020-05-14 02:01
From	"Dr. Richard Boulanger"
Subject	Re: [Csnd] Machine learning for composing -- where are we at?
	WOW. We need these opcodes in Csound! To write Csound lovesongs in "all" styles. Barry's Vercoe's "muse". _____________________________________________ Dr. Richard Boulanger Professor of Electronic Production and Design Professional Writing and Music Technology Division Berklee College of Music ______________________________________________ OFFICE: 1126 Boylston St., Suite 201 (EPD), Suite 208 (Boulanger) ______________________________________________ President of Boulanger Labs - http://boulangerlabs.com Author & Editor of The Csound Book - http://mitpress.mit.edu/books/csound-book Author & Editor of The Audio Programming Book - http://mitpress.mit.edu/books/audio-programming-book ______________________________________________ about: http://www.boulangerlabs.com/#about about: http://www.csounds.com/community/developers/dr-richard-boulanger/ music: http://www.csounds.com/community/developers/dr-richard-boulanger/dr-richard-boulanger-music/ ______________________________________________ email: rboulanger@berklee.edu facebook: https://www.facebook.com/richard.boulanger.58 On Fri, May 8, 2020 at 10:21 AM Michael Gogins <michael.gogins@gmail.com> wrote: TL;DR: Can generate songs up to minutes in length -- as raw waveforms. With vocals, lyrics, instruments, and all. Audio quality is not up to snuff. But still, VERY VERY IMPRESSIVE. This, insofar as my limited expertise in this domain is concerned, is one of several projects (like Magenta) that represents the current state of the art in the use of artificial intelligence/machine learning methods to generate music from scratch. A snapshot (I would guess out of date) of the code is at https://github.com/openai/jukebox. The repository has a link to a paper which explains how it works, to sample pieces, and to a colab notebook. This is a company, a very interesting company (https://openai.com/about/). And the software license reflects that. I, as a composer and programmer, could freely use this software, modify this software, and publish music made using this software -- BUT NOT IF I EVER WANT TO MAKE ONE RED CENT OFF MY MUSIC. Presumably I could pay for a license, but there's no obvious way to do that. Maybe that's just for the investors in this company. That's too bad, but this is still very interesting, and it should be possible (but maybe not easy) to create similar software based on the ideas in this software that would be truly open source. The lead author is clearly a live wire. Regards, Mike ----------------------------------------------------- Michael Gogins Irreducible Productions http://michaelgogins.tumblr.com Michael dot Gogins at gmail dot com Csound mailing list Csound@listserv.heanet.ie https://listserv.heanet.ie/cgi-bin/wa?A0=CSOUND Send bugs reports to https://github.com/csound/csound/issues Discussions of bugs and features can be posted here Csound mailing list Csound@listserv.heanet.ie https://listserv.heanet.ie/cgi-bin/wa?A0=CSOUND Send bugs reports to https://github.com/csound/csound/issues Discussions of bugs and features can be posted here