Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MotionBeat: Motion-Aligned Music Representation via Embodied Contrastive Learning and Bar-Equivariant Contact-Aware Encoding

Xuanchen Wang, Heng Wang, Weidong Cai

TL;DR

MotionBeat tackles the gap between audio-only music representations and embodied rhythm by learning motion-aligned music embeddings. It introduces Embodied Contrastive Loss ($L_{ECL}$) with tempo-aware negatives and beat-jitter negatives, and Structural Rhythm Alignment Loss ($L_{SRAL}$) using Soft-DTW and Earth Mover’s Distance to align beat/bar rhythm with motion. Architectural innovations—Bar-Equivariant Phase Rotations and Contact-Guided Attention—explicitly model cyclic rhythm and emphasize motion-synchronized events. Empirical results show MotionBeat outperforms strong baselines on music-to-dance generation and transfers effectively to beat tracking, tagging, genre/instrument classification, emotion recognition, and cross-modal retrieval, highlighting motion as a powerful supervisory signal for music representation.

Abstract

Music is both an auditory and an embodied phenomenon, closely linked to human motion and naturally expressed through dance. However, most existing audio representations neglect this embodied dimension, limiting their ability to capture rhythmic and structural cues that drive movement. We propose MotionBeat, a framework for motion-aligned music representation learning. MotionBeat is trained with two newly proposed objectives: the Embodied Contrastive Loss (ECL), an enhanced InfoNCE formulation with tempo-aware and beat-jitter negatives to achieve fine-grained rhythmic discrimination, and the Structural Rhythm Alignment Loss (SRAL), which ensures rhythm consistency by aligning music accents with corresponding motion events. Architecturally, MotionBeat introduces bar-equivariant phase rotations to capture cyclic rhythmic patterns and contact-guided attention to emphasize motion events synchronized with musical accents. Experiments show that MotionBeat outperforms state-of-the-art audio encoders in music-to-dance generation and transfers effectively to beat tracking, music tagging, genre and instrument classification, emotion recognition, and audio-visual retrieval. Our project demo page: https://motionbeat2025.github.io/.

MotionBeat: Motion-Aligned Music Representation via Embodied Contrastive Learning and Bar-Equivariant Contact-Aware Encoding

TL;DR

MotionBeat tackles the gap between audio-only music representations and embodied rhythm by learning motion-aligned music embeddings. It introduces Embodied Contrastive Loss () with tempo-aware negatives and beat-jitter negatives, and Structural Rhythm Alignment Loss () using Soft-DTW and Earth Mover’s Distance to align beat/bar rhythm with motion. Architectural innovations—Bar-Equivariant Phase Rotations and Contact-Guided Attention—explicitly model cyclic rhythm and emphasize motion-synchronized events. Empirical results show MotionBeat outperforms strong baselines on music-to-dance generation and transfers effectively to beat tracking, tagging, genre/instrument classification, emotion recognition, and cross-modal retrieval, highlighting motion as a powerful supervisory signal for music representation.

Abstract

Music is both an auditory and an embodied phenomenon, closely linked to human motion and naturally expressed through dance. However, most existing audio representations neglect this embodied dimension, limiting their ability to capture rhythmic and structural cues that drive movement. We propose MotionBeat, a framework for motion-aligned music representation learning. MotionBeat is trained with two newly proposed objectives: the Embodied Contrastive Loss (ECL), an enhanced InfoNCE formulation with tempo-aware and beat-jitter negatives to achieve fine-grained rhythmic discrimination, and the Structural Rhythm Alignment Loss (SRAL), which ensures rhythm consistency by aligning music accents with corresponding motion events. Architecturally, MotionBeat introduces bar-equivariant phase rotations to capture cyclic rhythmic patterns and contact-guided attention to emphasize motion events synchronized with musical accents. Experiments show that MotionBeat outperforms state-of-the-art audio encoders in music-to-dance generation and transfers effectively to beat tracking, music tagging, genre and instrument classification, emotion recognition, and audio-visual retrieval. Our project demo page: https://motionbeat2025.github.io/.
Paper Structure (16 sections, 9 equations, 1 figure, 3 tables)

This paper contains 16 sections, 9 equations, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Training Objective
  4. Embodied Contrastive Loss
  5. Structural Rhythm Alignment Loss
  6. Overall Loss
  7. Model Architecture
  8. Input Representation
  9. Bar-Equivariant Phase Rotations
  10. Contact-Guided Attention
  11. Experiments and Results
  12. Dataset and Implementation Details
  13. Downstream Tasks
  14. Baseline Methods
  15. Result Analysis
  16. ...and 1 more sections

Figures (1)

  • Figure 1: Architecture of MotionBeat. Audio and motion inputs are converted into beat-synchronous tokens and passed through encoders with bar-equivariant phase rotations and contact-guided attention. Embeddings are trained with the Embodied Contrastive Loss (ECL), while auxiliary rhythm heads provide onset envelopes and contact pulses for the Structural Rhythm Alignment Loss (SRAL). The total loss integrates both objectives to learn motion-aligned music representations.