TMD-Bench: A Multi-Level Evaluation Paradigm for Music-Dance Co-Generation

2026-05-03

AI Summary

Existing approaches struggle to effectively evaluate rhythmic coupling and cross-modal alignment in music-and-dance generation. This work proposes a multi-level evaluation framework that systematically assesses text-driven joint generation systems along three dimensions: single-modality quality, adherence to textual instructions, and music–dance rhythmic alignment. Innovatively integrating physically computable metrics with human perceptual judgments, the study introduces the first rhythm alignment dataset and a structured music semantic descriptor, alongside a unified baseline model named RhyJAM. Experiments reveal that prevailing audio-visual models exhibit notable deficiencies in beat synchronization, whereas RhyJAM significantly enhances beat-level cross-modal alignment while maintaining high-quality unimodal outputs.

Abstract

Unified audio-visual generation is rapidly gaining industrial and creative relevance, enabling applications in virtual production and interactive media. However, when moving from general audio-video synthesis to music-dance co-generation, the task becomes substantially harder: musical rhythm, phrasing, and accents must drive choreographic motion at fine temporal resolution, and such rhythmic coupling is not captured by unimodal metrics or generic audiovisual consistency scores used in current evaluation practice. We introduce TMD-Bench, a benchmark for text-driven music-dance co-generation that assesses systems across unimodal generation quality, instruction adherence, and cross-modal rhythmic alignment. The benchmark integrates computable physical metrics with perceptual multimodal judgments, and is supported by a curated rhythm-aligned music-dance dataset and a fine-grained Music Captioner for structured music semantics. TMD-Bench further reveals that (i) modern commercial audio-visual models, such as Veo 3 and Sora 2, produce high-quality music and video, while rhythmic coupling remains less consistently optimized and leaves room for improvement, and (ii) our unified baseline RhyJAM trained on rhythm-aligned data achieves competitive beat-level synchronization while maintaining competitive unimodal fidelity. This presents prospects for building next-generation music-dance models that explicitly optimize rhythmic and kinetic coherence.

Problem

• music-dance co-generation
• rhythmic alignment
• cross-modal synchronization
• evaluation benchmark
• audio-visual generation

Innovation

• music-dance co-generation
• rhythmic alignment
• multimodal evaluation
• TMD-Bench
• beat-level synchronization