# MERIT: Learning Disentangled Music Representations for Audio Similarity

Source: [MERIT: Learning Disentangled Music Representations for Audio Similarity](<https://www.sota2.com/research/paper/6a167adadbd7150506f80322>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-06T06:00:02.123Z

Authors: Abhinaba Roy, Junyi Liang, Dorien Herremans

Published: 2026-05-26T17:49:18.000Z

## Abstract

Current music similarity models typically compute a single, monolithic score, entangling distinct musical dimensions like melody, rhythm, and timbre. This limits user control and interpretability, making it impossible to execute nuanced queries. We introduce MERIT, a framework for learning disentangled, factor-specific music representations tailored to these three core dimensions. To overcome the lack of isolated musical variations in real-world audio, we use a novel training strategy that uses conditional audio generation and source-separated stems to strongly encourage single-factor variation in training data. Our evaluations demonstrate strong factor-wise disentanglement. Each head responds strongly to its intended perceptual dimension while remaining near chance on the others, a representational property that holds across both the synthetic training domain and independent real-world audio.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Cover identification on Covers80](<https://www.sota2.com/research/sota/cover-identification-on-covers80.md>): Cover identification; Covers80; Accuracy: 69.9.

- [Rhythmic groove selectivity on Ballroom](<https://www.sota2.com/research/sota/rhythmic-groove-selectivity-on-ballroom.md>): Rhythmic groove selectivity; Ballroom; Accuracy: 88.

- [Instrument-class selectivity on MUSDB18-HQ](<https://www.sota2.com/research/sota/instrument-class-selectivity-on-musdb18-hq.md>): Instrument-class selectivity; MUSDB18 HQ; Accuracy: 78.9.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.27346>)

[PDF](<https://arxiv.org/pdf/2605.27346.pdf>)

[Implementation](<https://github.com/AMAAI-Lab/MERIT>)