Research map / Vision-language and multimodal / Cross-modal learning
VoViT: Low Latency Graph-Based Audio-Visual Voice Separation Transformer
Juan F. Montesinos, Venkatesh S. Kadandale, Gloria Haro
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 lead voice enhancer stage 2 (r=1) (ablation) | 10-block AV ST-transformer · Voxceleb2 wild test | SDR: 9.68 -> 10.03 | Improved |
| add lead voice enhancer stage 2 (r=1) (ablation) | 10-block AV ST-transformer · Voxceleb2 wild test | SIR: 15.75 -> 18.18 | Improved |
| run enhancer twice (r=2) instead of once (ablation) | VoViT · Voxceleb2 wild test | SDR: 10.03 -> 9.78 | Hurt |
| run enhancer twice (r=2) instead of once (ablation) | VoViT · Voxceleb2 wild test | SIR: 18.18 -> 19.09 | Improved |
| use larger Visual Voice U-Net instead of Y-Net U-Net as enhancer (ablation) | VoViT · Voxceleb2 wild test | SDR: 10.03 -> 10.05 | No clear effect |
| use 18-block instead of 10-block single-stage transformer (ablation) | VoViT-s1 · Voxceleb2 wild test | SDR: 9.68 -> 9.27 | Hurt |
| feed audio-visual signal to encoder only instead of video to encoder and audio to decoder (ablation) | transformer · Acappella unseen-unheard | SDR: 8.64 -> 10.63 | Improved |
| feed audio-visual signal only instead of also giving clean audio to decoder (ablation) | transformer · Acappella unseen-unheard | SDR: 9.98 -> 10.63 | Improved |
| replace Y-Net with AV ST-transformer | VoViT · Acappella unseen-unheard | SDR: 6.41 -> 10.63 | Improved |
| replace Visual Voice with VoViT | VoViT · Voxceleb2 wild test | SIR: 16.11 -> 18.18 | Improved |
| replace Visual Voice with VoViT | VoViT · Voxceleb2 wild test | SDR: 9.92 -> 10.03 | No clear effect |
| train on speech instead of singing voice | VoViT-s1 (4 blocks) · Acappella English two-singer mixtures | SDR: 14.85 -> 3.89 | Hurt |
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.
- Visual Voice — VisualVoice: Audio-Visual Speech Separation With Cross-Modal Consistency (CVPR 2021) · better on 4 of 4 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
- VisualVoice: Audio-Visual Speech Separation With Cross-Modal Consistency (CVPR 2021 · similarity 0.88)
- RTFS-Net: Recurrent Time-Frequency Modelling for Efficient Audio-Visual Speech Separation (ICLR 2024 · similarity 0.86)
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.84)
- Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation (NeurIPS 2024 · similarity 0.84)
- TIGER: Time-frequency Interleaved Gain Extraction and Reconstruction for Efficient Speech Separation (ICLR 2025 · similarity 0.84)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.83)
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.
