MaskBeat: Loopable Drum Beat Generation

TL;DR

MaskBeat is a transformer-based model that generates loopable drum patterns using bidirectional attention and custom loss functions, resulting in more coherent and higher quality drum sequences.

Contribution

It introduces a novel transformer architecture with bidirectional attention and custom loss functions specifically designed for drum pattern generation.

Findings

Outperforms baseline methods in quality and coherence

Generates drum patterns with improved musical relevance

Uses parallel instrument generation for efficiency

Abstract

We present MaskBeat, a transformer-based approach for loopable drum pattern generation. Rather than predicting drum hits sequentially, our method uses bidirectional attention with iterative refinement, allowing instruments to be generated in parallel while maintaining musical coherence. Additionally, we introduce custom loss functions that capture drum-specific musical relationships. Our experiments show that MaskBeat generates higher quality and more musically coherent drum patterns than baseline approaches.