Research map / Image generation / Autoregressive and latent generation
SyncTrack: Rhythmic Stability and Synchronization in Multi-Track Music Generation
Hongrui Wang, Fan Zhang, Zhiyuan Yu, Ziya Zhou, Xi Chen, Can Yang, Yang Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add track-specific module (learnable instrument prior) to the backbone (ablation) | SyncTrack backbone · Slakh2100 mixture | FAD: 2.570 -> 1.742 | Improved |
| add track-specific module to the backbone (ablation) | SyncTrack backbone · Slakh2100 piano track | FAD: 6.170 -> 1.695 | Improved |
| add global cross-track attention on top of track-specific module (ablation) | Backbone w/ track-specific · Slakh2100 mixture | FAD: 1.742 -> 1.627 | Improved |
| add time-specific cross-track attention on top of track-specific module (ablation) | Backbone w/ track-specific · Slakh2100 mixture | FAD: 1.742 -> 1.429 | Improved |
| add time-specific attention after global attention (full model) (ablation) | Backbone w/ track-specific + global attention · Slakh2100 mixture | FAD: 1.627 -> 1.260 | Improved |
| add global attention after time-specific attention (full model) (ablation) | Backbone w/ track-specific + time-specific attention · Slakh2100 mixture | FAD: 1.429 -> 1.260 | Improved |
| apply global then time-specific attention in sequence instead of reversed order (ablation) | SyncTrack · Slakh2100 mixture | FAD: 1.681 -> 1.260 | Improved |
| use both attentions in every module instead of alternating one per module (ablation) | SyncTrack · Slakh2100 mixture | FAD: 1.586 -> 1.260 | Improved |
| replace joint-distribution architecture (MSG-LD) with shared/specific split (SyncTrack) | multi-track generator · Slakh2100 mixture | FAD: 1.31 -> 1.26 | Improved |
| SyncTrack vs MSG-LD on cross-track beat synchronization | multi-track generator · Slakh2100 mixture | CBS: 0.3861 -> 0.5206 | Improved |
| SyncTrack vs MSG-LD on beat dispersion | multi-track generator · Slakh2100 mixture | CBD (mean): 0.3714 -> 0.2681 | Improved |
| generate fewer tracks given known tracks (bass only vs all four) (ablation) | SyncTrack conditional generation · Slakh2100 conditional | CBS: 0.5206 -> 0.5620 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- JEN-1 Composer — JEN-1 Composer: A Unified Framework for High-Fidelity Multi-Track Music Generation (AAAI 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MoMu-Diffusion: On Learning Long-Term Motion-Music Synchronization and Correspondence (NeurIPS 2024 · similarity 0.87)
- JEN-1 Composer: A Unified Framework for High-Fidelity Multi-Track Music Generation (AAAI 2025 · similarity 0.86)
- SongBloom: Coherent Song Generation via Interleaved Autoregressive Sketching and Diffusion Refinement (NeurIPS 2025 · similarity 0.86)
- MIDI-GPT: A Controllable Generative Model for Computer-Assisted Multitrack Music Composition (AAAI 2025 · similarity 0.85)
- Structured Multi-Track Accompaniment Arrangement via Style Prior Modelling (NeurIPS 2024 · similarity 0.85)
- Unifying Symbolic Music Arrangement: Track-Aware Reconstruction and Structured Tokenization (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
