An event-based sequence modeling approach to recognizing non-triad chords with oversegmentation minimization

TL;DR

This paper presents a segment-level sequence-to-sequence approach for automatic chord recognition, effectively addressing oversegmentation and recognizing complex non-triad chords with improved accuracy.

Contribution

It introduces a novel segment-based prediction framework, structured tokenization, and pre-training methods tailored for time-aligned chord modeling, enhancing recognition of complex chords.

Findings

Improved chord recognition accuracy, especially for complex non-triad chords

Reduced oversegmentation errors in chord boundary detection

Effective use of structured tokenization and pre-training for chord modeling

Abstract

Automatic chord recognition (ACR) extracts time-aligned chord labels from music audio recordings. Despite recent advances, ACR still struggles with oversegmentation, data scarcity, and imbalance, especially in recognizing complex chords such as non-triads, which are unpopular in existing datasets. To address these challenges, we reformulate ACR as a segment-level sequence-to-sequence prediction task, where chord sequences are predicted auto-regressively rather than frame by frame. This design mitigates excessive segmentation by detecting chord changes only at segment boundaries. We further introduce two types of token representations and an encoder pre-training method, both specifically designed for time-aligned chord modeling. Experimental results show that our model improves performance in both chord recognition and segmentation, with notable gains for complex and infrequent chord types. These findings demonstrate the effectiveness of segment-level sequence modeling, structured tokenization, and representation learning for advancing chord recognition systems.