Coconet

Coconet is an instance of an orderlessNADE, that uses deep convolutional neural networks to perform music inpaintings through Gibbs sampling. In the demo below, we call infill() to create Bach-style 4-part harmonies with the provided input, and display the result.

Infilling the first voice

The default behaviour of the model is to infill all of the "silence" in a NoteSequence. In this example, only the first voice (Soprano) is present for all the time steps, so Coconet will infill the other 3 voices (Alto, Tenor and Bass)

Input:

Result: Loading...

It took:

Infilling the third voice with configuration parameters.

In this example, only the Tenor voice is present, so we infill the first 2 and last voices (Soprano, Alto, Bass). In addition, we also use the temperature and numIterations parameters to the infill() method.

Input:

Result: Loading...

It took:

Infilling a missing section from all the voices

As mentioned before, the default behaviour of the model is to infill all of the "silent" measures. In order to distinguish between intended rests and infill sections, you can use the infillMask parameter to specify exactly which section to be infilled.

Input:

Result: Loading...

It took:

Total Leaked Memory: