The Drum Machine of Tao

Xiaowan Yi; Mathieu Barthet

Image credit: Xiaowan Yi; Mathieu Barthet

• Format: poster
• Session: posters-3
• Paper PDF link
• Presence: in person
• Duration: 5
• Type: short

Abstract:

The Drum Machine of Tao (Tao) is a machine learning–based system that reverse-engineers sequencer parameters and one-shot percussive samples from drum loops, restoring low-level editability to sampled loops that would otherwise be frozen in audio waveforms. The philosophy behind this system is inspired from Taoism: that which returns to its primal state is the great Way of Tao. In this paper, we present the system design of Tao, which includes a state-of-the-art drum source separation model, a sequencer parameter estimation model, and a bespoke one-shot sample extraction algorithm that leverages differentiable audio synthesis. Results from a prototype are available for listening.