Gearslutz

Gearslutz (https://www.gearslutz.com/board/)
-   Music Computers (https://www.gearslutz.com/board/music-computers/)
-   -   Amazon Announces 'AWS DeepComposer' (https://www.gearslutz.com/board/music-computers/1290083-amazon-announces-aws-deepcomposer.html)

Rtroost 3rd December 2019 04:46 PM

Amazon Announces 'AWS DeepComposer'
 
Check it out: announced at AWS reinvent this week.

https://aws.amazon.com/deepcomposer/faq/

It includes a MIDI keyboard and cloud service.

Will me interesting to see what it actually does....

Sarah Connor?

vitocorleone123 3rd December 2019 07:49 PM

It's definitely a step up from their Shallow Composer bot that posts on GS all day instead of making music.

jummpp

mmahan40 3rd December 2019 07:52 PM

Quote:

Originally Posted by Rtroost (Post 14360793)
Check it out: announced at AWS reinvent this week.

https://aws.amazon.com/deepcomposer/faq/

It includes a MIDI keyboard and cloud service.

Will me interesting to see what it actually does....

Sarah Connor?

Pretty cool, but I'm also skeptical that music will turn out to be a great application for AI. I say that mostly because I would argue that good music is a result of good taste, and not necessarily a result of understanding music theory.

I'm sure there will eventually be neural networks that produce great music, but is that really what people are going to want? I certainly don't know, but I think music is an essential part of the human experience, and I'd like to think it's the humanity that connects artists/composers and listeners, just as much, if not more than the music itself.

I also think AI in music brings up a lot of legal questions. General Adversarial Networks (GANS) are one thing since they can potentially be trained without using existing works, but any convolution or recurrent neural network would have to be trained on existing works. This begs the question, if a network is trained on said works, who owns the output of the network? I would argue the original composer should own at least some of it, since the process by which the network is learning and generating ideas is very similar to sampling or interpolation.

That said I think GANS will be very useful for things like playing chords and then the network will add in a melody in real time, as opposed to generating entire pieces.

Rtroost 3rd December 2019 11:15 PM

Yeah I share your skepticism. We’ve yet to see anything in art, fiction, or news reporting from ML that is compelling. But it’ll happen.

Money for lawyers for sure.

I’m still keen to play with it.

Rtroost 6th December 2019 07:45 PM

Word from my colleagues that attended the sessions is that its pretty underwhelming.

mattiasnyc 6th December 2019 08:06 PM

Quote:

Originally Posted by mmahan40 (Post 14361154)
This begs the question, if a network is trained on said works, who owns the output of the network?

This.

Given how the fine print has written for some other social media apps where the service provider gains great rights to user uploaded content like photos one wonders where this will lead.

I say we need to establish a new model for art and economics. If not the bulk of those making at least some money off of it will get screwed.