Research map / Video / Video generation
Towards Streaming Synchronized Spatial Audio Generation via Autoregressive Diffusion Transformer
Ke Lei, Yu Zhang, Changhao Pan, Xueyi Pu, Wenxiang Guo, Ruiqi Li, Zhou Zhao
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 |
|---|---|---|---|
| use VideoMAE encoder with SVAC contrastive learning instead of frame-level CLIP features (ablation) | SwanSphere video encoder · video-to-spatial audio | FD / angular error: FD 140.28 -> 120.28; angular 1.34 -> 1.03 | Improved |
| add physics-based negatives (temporal offset, audio and video rotation) to semantic negatives in SVAC (ablation) | SVAC · video-to-spatial audio | angular error: 1.12 -> 1.03 | Improved |
| add multi-objective ODPO fine-tuning (ablation) | SwanSphere-L · video-to-spatial audio | FD / angular error: FD 133.91 -> 120.28; angular 1.22 -> 1.03 | Improved |
| add history conditioning via history encoder (ablation) | SwanSphere · video-to-spatial audio | FD: 128.15 -> 120.28 | Improved |
| scale model from 0.43B to 1.09B (ablation) | SwanSphere · video-to-spatial audio | FD / angular error: FD 139.81 -> 120.28; angular 1.33 -> 1.03 | Improved |
| use causal autoregressive diffusion (LocDiT) instead of a global DiT (ablation) | SwanSphere · video-to-spatial audio | first-chunk latency / FD: 6.47s -> 0.21s; FD 123.08 -> 120.28 | Improved |
| use global DiT instead of autoregressive diffusion (ablation) | SwanSphere · video-to-spatial audio | elevation error: 1.14 -> 0.91 (elevation error) | Mixed |
| add caption conditioning to panoramic video (ablation) | SwanSphere · caption test subset | FD / angular error: FD 120.28 -> 118.31; angular 1.03 -> 0.96 | 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.
- OmniAudio — OmniAudio: Generating Spatial Audio from 360-Degree Video (ICML 2025) · better on 5 of 5 shared results
- ViSAGe — VISAGE: Video Instance Segmentation with Appearance-Guided Enhancement (ECCV 2024) · 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
- Both Ears Wide Open: Towards Language-Driven Spatial Audio Generation (ICLR 2025 · similarity 0.88)
- ViSAGe: Video-to-Spatial Audio Generation (ICLR 2025 · similarity 0.86)
- OmniAudio: Generating Spatial Audio from 360-Degree Video (ICML 2025 · similarity 0.86)
- MRSAudio: A Large-Scale Multimodal Recorded Spatial Audio Dataset with Refined Annotations (NeurIPS 2025 · similarity 0.86)
- AV-Cloud: Spatial Audio Rendering Through Audio-Visual Cloud Splatting (NeurIPS 2024 · similarity 0.85)
- Audible World Models: Spatially Aware Sound Generation for 3D Worlds (NeurIPS 2026 · similarity 0.85)
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.
