Choose a song. In order to play it, you might have to turn off silent mode in your phone.
Train the song. The Machine Learning model MidiMe is used in order to train the song. In order for it to work in iOS, WebGL must be turned on in the browser settings.
With the model trained, you may now use it to produce original music samples similar to the trained song.
Audio is managed with Tone.js. Machine Learning is performed using the Tensorflow library Magenta.js. Music notation is rendered using VexFlow.