Old 03-24-2019, 12:07 AM   #1
lolilol1975
Human being with feelings
 
Join Date: Dec 2015
Posts: 1,739
Default Google Magenta

Everyone has seen the Google Doodle that hilariously failed at harmonizing like Bach.

Still the technology behind it is avalaible and demonstrated here. Some of these apps are pretty "clever" !
Performance RNN is pretty cool and can input and output MIDI for live music generation !
Nsynth generates sounds that you'd have a hard time generating with a standard synth, especially when you combine an instrument with a natural sound.

Is someone going to try to make plugins that use Magenta for Reaper ? I can imagine so many possible uses, either for MIDI (melodic lines, drum beats, humanizer, etc), or for working on audio (an intelligent compressor for instance ?)

Last edited by lolilol1975; 03-24-2019 at 12:38 AM.
lolilol1975 is offline   Reply With Quote
Old 03-24-2019, 01:38 PM   #2
chip mcdonald
Human being with feelings
 
chip mcdonald's Avatar
 
Join Date: May 2006
Location: NA - North Augusta South Carolina
Posts: 4,294
Default

How did it fail?
__________________
]]] guitar lessons - www.chipmcdonald.com [[[
WEAR A FRAKKING MASK!!!!
chip mcdonald is offline   Reply With Quote
Old 03-25-2019, 04:01 AM   #3
martifingers
Human being with feelings
 
Join Date: May 2011
Posts: 2,694
Default

This is a great example of how to use this sort of aid I think:
https://www.youtube.com/watch?v=hOvxchH9uQc

And (of course) Adam Neely has a very good analysis of it here:
https://www.youtube.com/watch?v=xDqx14lZ_ls
martifingers is offline   Reply With Quote
Old 03-25-2019, 10:25 AM   #4
chip mcdonald
Human being with feelings
 
chip mcdonald's Avatar
 
Join Date: May 2006
Location: NA - North Augusta South Carolina
Posts: 4,294
Default

I agree somewhat with Neely's analysis (the lick "harmonization" was ... curious), except that the choices that are "right" are subjective and have to be within a context. The ML algorithm has no context, and was perhaps trained on passages that are motifs from different parts of an arrangement.

In that context I think it did very well. When I played with it I thought about 1 in 3 trials produced something I felt was "ugly", and maybe 1 in.. 7 or 8 trials something I thought was "interesting" or produced an emotional response.

The ML is not an algorithm based on baroque/classical arrangement rules, so I would not expect it to to adhere stringently to it. I would think it would almost be a trivial matter for them to add a filter that weeds out "non-authentic" violations.

I think that would negate the point of ML/GAN approaches, though, since the "rules" are subjective and there has already been plenty of "stochastic input iterative rules based engine programs" that churn out more non-contextural passages that may adhere to classic harmonic movement better, but are never going to literally "create" something new. A dead end.
__________________
]]] guitar lessons - www.chipmcdonald.com [[[
WEAR A FRAKKING MASK!!!!

Last edited by chip mcdonald; 03-25-2019 at 10:37 AM.
chip mcdonald is offline   Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -7. The time now is 06:51 AM.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.