Make unusual new sounds with machine learning.
Clone or download
Latest commit 3f0fa96 May 24, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
images initial code commit May 23, 2017
sounds initial sound commit May 23, 2017
src initial code commit May 23, 2017
style initial code commit May 23, 2017
third_party initial code commit May 23, 2017
.gitignore initial code commit May 23, 2017
CONTRIBUTING.md initial code commit May 23, 2017
LICENSE initial code commit May 23, 2017
README.md adding credits May 24, 2017
index.html initial code commit May 23, 2017
package.json initial code commit May 23, 2017
webpack.config.js initial code commit May 23, 2017

README.md

This is the source code and sounds to the A.I. Experiment, NSynth: Sound Maker.

This experiment lets you play with new sounds created with machine learning. It's built using Nsynth, a research project that trained a neural network on over 300,000 instrument sounds.

NSynth lets you select and combine two sound sources. Use the slider to change the interpolation between the two sound sources.

CREDITS

Built by Yotam Mann with friends on the Magenta and Creative Lab teams at Google. It uses TensorFlow, Tone.js and open-source tools from the Magenta project. Check out more at A.I. Experiments.

Building

To build and run your own, first make sure that you have Node.js and webpack installed.

To install the dependencies and build the project, in the terminal run:

npm install
webpack -p

You can then run a local server to play the site.

Audio files

All of the sound combinations can be found in the sounds folder. The files are named with the two instrument sounds and an underscore in between. Inside the file you will find 35 mp3 files. The mp3s are named with their interpolation point (0-4), then an underscore, then their midi note value.

For example, Bass_Trombone_Solo_A_bike_bell/1_68.mp3 is a combination of Bass_Trombone_Solo_A and bike bell. Since the file is called 1_68.mp3, that means it's the second interpolation point between trombone and bike bell (25% trombone and 75% bell) and the note value is 68 (G#5).

The samples aren't always perfectly in tune and they do sound a bit unusual – but we think that’s what makes them fun.

Keyboard

You can play the keyboard by touching or clicking the keys or plug in a MIDI keyboard. The center row of your computer keyboard will also trigger the keys.