Make some noise and record your voice! This is a Neural Beatbox, made with RNN-based Rhythm Generation and Audio Classification. It uses TensoFlow.js, Magenta.js and p5.js.
The rhythm generation part was based on Magenta.js’s Drum RNN and @teropa’s Neural Drum Machine.
Creator Nao Tokui used his own Keras model and converted it to a TensorFlow.js model to classify sound segments.
The deep learning models analyze and build a drumkit then generate drum patterns infinitely with your sound inputs. Try it out here!
This currently only works in Chrome.