Device Details

Device Overview

Name/Version: deeploops 1.1
Author: aiton  
Description: deeploops is a drum pattern generator using Artificial Inteligence. It creates a MIDI clip of 2 bars length. Drag deeploops.amxd on a drum track, and press on the button to create a clip. The clip contains MIDI notes for a typical Ableton drum rack. You can set how much "chaos" is in the notes. 1 is a good value to start with. With smaller values, the generated drum patterns might not vary much. With higher values, there is more randomness in the notes.

You can influence the generated notes, by creating a primer clip on the drum track. deeploops will take the notes in the primer clip and try to continue the sequence in a musically meaningful way. Check out the included demo set for 2 examples of the primer. The primer clip should have the name "primer" to be recognized as such.

deeploops comes with a pre-trained deep neural network. The trained model is strongly based on Google Magentas drums_rnn:

after unzip, 2 lines are to be added to C:\Program Files\Cycling '74\Max 7
esources\packages\max-mxj\java-classes\, or wherever M4L is installed. First save a copy of the original file, and then open it with a text editor and add the following two lines:

max.dynamic.jar.dir [deeploops-root]/java/lib
max.jvm.option -Djava.library.path=[deeploops-root]/bin

where [deeploops-root] is the folder containing deeploops.amxd. Use slashes '/' instead of backslashes e.g. c:/mywork/deeploops. Restart Ableton.

- PC only (Windows 7 or higher; 64-bit, x86). It comes with the tensorflow dll, see Porting it to MacOS should be no big issue, but unfortunately I have no Mac to test. If anybody would like to help me here...
- Generated clips are quantized, and notes have all same length.
- Only one trained genre: EDM. If you have many MIDI drum loops of any other genre, please get in touch, and we might train it for another style of music.

v1.1: fixed issue with detecting path to pickle file
v1.0: initial revision

Device Details

Tags drum, utility
Live Version Used: 10.0.1
Max Version Used: 8
Date Added: Mar 23 2018 22:12:12
Date Last Updated: Apr 17 2018 06:58:57
Device Type: midi_device
Download URL (report link)
License (more info): AttributionNoDerivatives


A little demo how it sounds:

Seems the zip file is corrupt?

fixed that, thanks

It would be nice to have something like mutable instruments grids or some of the noise engineering stuff.
Apparently they analyzed patterns from songs in classic genre and tribal ones. Like african rythm and patterns done live by tribes.
Could the AI actually learn from some audio files you give them or does it have to be midi ?
It would be really dope

Never tried the original Grids, but I tested @mots M4L port. I think deeploops is quite different, it is more "creative", though arguably whether creativity in a machine can exist. It generates quite surprising ideas for drum patterns. The intention is that a musician can take them as inspiration for breaks.

Right now training needs MIDI files. I'm currently looking into AI methods to generate drum patterns from audio, but can't say yet that this will work reliable enough.

Hi - i get an error on the device that says it can't load the pickle file. Is this related to my java paths or something? Max settings seem correct, since I get different errors if I don't correctly fill out the directory paths specified in the readme.txt

I put the deeploops folder on another drive, other than root of C and everything seems to be ok now. Was having super weird issues after putting the folder into my User/.../Music too (log errors about finding some Maxglu commands)

there seems to be an issue in determining the path where the device is located... thanks for reporting that, i will look into it.
Great to hear that you got it to run nevertheless!

Fixed detection of data folder with v1.1

really interested in further development of this project btw. I would like to experiement with training itself to see what can be achieved so then i can generate albums and make lots of money.

I'm happy to help port to OS X and help provide hip hop midi files to train. Contact me (my user name [at] gmail)

Login to comment on this device.

[ browse device library ]