Beat Blender

This experiment lets you explore and create beats in a fun new way using machine learning. Just drag the circle or draw a path to discover beats. It's built using a neural network trained on over 3.8 million drum beats. Using TensorFlow.js, the neural network runs locally in your web browser.

Beat Blender was first presented at the 2017 NIPS conference. Built by Torin Blankensmith and Kyle Phillips from the Creative Lab and Adam Roberts from the Magenta team at Google using MusicVAE.js and TensorFlow.js. Get the open source code on Github here.