Talking Drums: Generating drum grooves with neural networks

TL;DR

This paper introduces a neural network-based method for generating full drum kit patterns from kick-drum sequences, exploring sampling techniques and style-dependent consistency in output quality.

Contribution

It adapts a sequence-to-sequence neural network model from language translation to drum pattern generation and evaluates sampling techniques for improved musical output.

Findings

Sampling from the top three probable outputs improves pattern quality

Output consistency varies significantly across musical styles

The method effectively generates full drum kit patterns from kick sequences

Abstract

Presented is a method of generating a full drum kit part for a provided kick-drum sequence. A sequence to sequence neural network model used in natural language translation was adopted to encode multiple musical styles and an online survey was developed to test different techniques for sampling the output of the softmax function. The strongest results were found using a sampling technique that drew from the three most probable outputs at each subdivision of the drum pattern but the consistency of output was found to be heavily dependent on style.