BACHI: Boundary-Aware Symbolic Chord Recognition Through Masked Iterative Decoding on Pop and Classical Music
Mingyang Yao, Ke Chen, Shlomo Dubnov, Taylor Berg-Kirkpatrick
2025-10-08
Summary
This paper focuses on automatically figuring out the chords in music, but it's not just about analyzing the sound itself. It also looks at how to read chords from musical scores, and tries to make the computer 'think' about music more like a musician would.
What's the problem?
Current computer programs that identify chords are really good at working with actual audio recordings, but they haven't been developed much for reading chords directly from sheet music. This is because there isn't a lot of labeled sheet music data available for training these programs. Also, most programs don't break down the chord-finding process into steps that match how humans learn to identify chords – like first finding where a chord starts and ends, then figuring out the specific notes within it.
What's the solution?
The researchers created a better dataset of sheet music called POP909-CL, which has accurate chord labels and timing information. Then, they built a new computer model called BACHI that identifies chords in a way that mimics human musicians. BACHI first figures out where chords begin and end, and then separately determines the root note, the type of chord (major, minor, etc.), and how the chord is positioned (inversion).
Why it matters?
This work is important because it improves chord recognition for both audio and sheet music, and it does so by making the computer's process more similar to how people understand music. This could lead to better music software for things like automatic accompaniment, music transcription, and music education.
Abstract
Automatic chord recognition (ACR) via deep learning models has gradually achieved promising recognition accuracy, yet two key challenges remain. First, prior work has primarily focused on audio-domain ACR, while symbolic music (e.g., score) ACR has received limited attention due to data scarcity. Second, existing methods still overlook strategies that are aligned with human music analytical practices. To address these challenges, we make two contributions: (1) we introduce POP909-CL, an enhanced version of POP909 dataset with tempo-aligned content and human-corrected labels of chords, beats, keys, and time signatures; and (2) We propose BACHI, a symbolic chord recognition model that decomposes the task into different decision steps, namely boundary detection and iterative ranking of chord root, quality, and bass (inversion). This mechanism mirrors the human ear-training practices. Experiments demonstrate that BACHI achieves state-of-the-art chord recognition performance on both classical and pop music benchmarks, with ablation studies validating the effectiveness of each module.