Research map / Video / Video generation
OmniSonic: Towards Universal and Holistic Audio Generation from Video and Text
Weiguo Pian, Saksham Singh Kushwaha, Zhimin Chen, Shijian Deng, Kai Wang, Yunhui Guo, Yapeng Tian
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 MoE Gating to fuse the three cross-attention outputs (ablation) | OmniSonic · UniHAGen-Bench | FAD: 6.12 -> 3.07 | Improved |
| add MoE Gating to fuse the three cross-attention outputs (ablation) | OmniSonic · UniHAGen-Bench | MKL: 5.25 -> 2.79 | Improved |
| add MoE Gating to fuse the three cross-attention outputs (ablation) | OmniSonic · UniHAGen-Bench | Mean AT/AV score: 15.94 -> 18.54 | Improved |
| add MoE Gating to fuse the three cross-attention outputs (ablation) | OmniSonic · UniHAGen-Bench | WER: 0.56 -> 0.14 | Improved |
| add MoE Gating to fuse the three cross-attention outputs (ablation) | OmniSonic · UniHAGen-Bench | DeSync: 1.23 -> 0.72 | Improved |
| suppress speech branch gate score in MoE Gating (in-the-wild samples) (ablation) | OmniSonic · UniHAGen-Bench | speech generation | Hurt |
| suppress off-screen environmental branch gate score (ablation) | OmniSonic · UniHAGen-Bench | off-screen sound presence | Hurt |
| joint text-video conditioning with TriAttn-DiT versus text-only VoiceLDM | OmniSonic · UniHAGen-Bench | FAD: 3.58 -> 3.07 | Improved |
| use only CLIP visual features instead of also Synchformer features for synchronization | OmniSonic · UniHAGen-Bench | DeSync: 0.72 vs 0.38 (HunyuanVideo-Foley) | Hurt |
| OmniSonic versus best baseline on environmental faithfulness and temporal MOS | OmniSonic · UniHAGen-Bench | MOS-T: 3.86 -> 4.29 | Improved |
Closest papers by meaning
- UniFlow-Audio: Unified Flow Matching for Audio Generation from Omni-Modalities (NeurIPS 2026 · similarity 0.91)
- VinTAGe: Joint Video and Text Conditioning for Holistic Audio Generation (CVPR 2025 · similarity 0.90)
- Read, Watch and Scream! Sound Generation from Text and Video (AAAI 2025 · similarity 0.89)
- SoundCTM: Unifying Score-based and Consistency Models for Full-band Text-to-Sound Generation (ICLR 2025 · similarity 0.88)
- OmniFlow: Any-to-Any Generation with Multi-Modal Rectified Flows (CVPR 2025 · similarity 0.88)
- Diverse and Aligned Audio-to-Video Generation via Text-to-Video Model Adaptation (AAAI 2024 · 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.
