Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Roy, Abhinaba, Liang, Junyi, Herremans, Dorien
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2605.27346
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913164641173504
author Roy, Abhinaba
Liang, Junyi
Herremans, Dorien
author_facet Roy, Abhinaba
Liang, Junyi
Herremans, Dorien
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27346
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MERIT: Learning Disentangled Music Representations for Audio Similarity
Roy, Abhinaba
Liang, Junyi
Herremans, Dorien
Sound
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.
title MERIT: Learning Disentangled Music Representations for Audio Similarity
topic Sound
url https://arxiv.org/abs/2605.27346