Research map / Video / Video generation
JAM-Flow: Joint Audio-Motion Synthesis with Flow Matching
Mingi Kwon, Joonghyuk Shin, Jaeseok Jeong, Jaesik Park, Youngjung Uh
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 temporal attention mask to Motion-DiT (half joint, frozen Audio-DiT) (ablation) | JAM-Flow · HDTF ablation | LSE-C: 4.64 -> 5.76 | Improved |
| add temporal attention mask to Motion-DiT (half joint, frozen Audio-DiT) (ablation) | JAM-Flow · HDTF ablation | LSE-D: 9.07 -> 8.22 | Improved |
| add temporal attention mask to Motion-DiT (half joint, frozen Audio-DiT) (ablation) | JAM-Flow · Ablation | WER: 7.25% -> 6.76% | Improved |
| add temporal attention mask (with Audio-DiT finetuning) (ablation) | JAM-Flow · Ablation | FID: 5.748 -> 5.662 | Improved |
| add temporal attention mask (with Audio-DiT finetuning) (ablation) | JAM-Flow · Ablation | LSE-D: 7.99 -> 7.73 | Improved |
| finetune Audio-DiT in stage 2 (no mask) (ablation) | JAM-Flow · HDTF ablation | LSE-C: 4.64 -> 6.44 | Improved |
| finetune Audio-DiT in stage 2 (with mask) (ablation) | JAM-Flow · HDTF ablation | LSE-C: 5.76 -> 6.45 | Improved |
| finetune Audio-DiT in stage 2 (with mask) (ablation) | JAM-Flow · Ablation | WER: 6.76% -> 7.28% | Hurt |
| finetune Audio-DiT in stage 2 (no mask) (ablation) | JAM-Flow · Ablation | WER: 7.25% -> 7.93% | Hurt |
| unfreeze Audio-DiT during joint training (ablation) | JAM-Flow · LibriSpeech-PC TTS | WER: 3.38% -> 4.91% | Hurt |
| use full 22 joint-attention blocks instead of half (11) (ablation) | JAM-Flow · HDTF ablation | LSE-C: 4.64 -> 4.81 | Mixed |
| use full 22 joint-attention blocks instead of half (11) (ablation) | JAM-Flow · Ablation | WER: 7.25% -> 6.88% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Hallo3 — Hallo3: Highly Dynamic and Realistic Portrait Image Animation with Video Diffusion Transformer (CVPR 2025) · better on 4 of 4 shared results
- VoiceCraft-Dub — VoiceCraft-Dub: Automated Video Dubbing with Neural Codec Language Models (ICCV 2025) · better on 2 of 2 shared results
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
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.89)
- BinauralFlow: A Causal and Streamable Approach for High-Quality Binaural Speech Synthesis with Flow Matching Models (ICML 2025 · similarity 0.88)
- Faces that Speak: Jointly Synthesising Talking Face and Speech from Text (CVPR 2024 · similarity 0.88)
- EmotiveTalk: Expressive Talking Head Generation through Audio Information Decoupling and Emotional Video Diffusion (CVPR 2025 · similarity 0.87)
- Synergizing Motion and Appearance: Multi-Scale Compensatory Codebooks for Talking Head Video Generation (CVPR 2025 · similarity 0.87)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.87)
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.
