Research map / Label-efficient and robust learning / Recognition and long-tail
Seeking the Shape of Sound: An Adaptive Framework for Learning Voice-Face Association
Peisong Wen, Qianqian Xu, Yangbangyan Jiang, Zhiyong Yang, Yuan He, Qingming Huang
Check 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 the full two-level alignment and adaptive identity reweighting framework | voice-face association learning · 1:2 matching, verification and retrieval | accuracy, AUC, and mAP | Improved |
| Add explicit modality alignment (ablation) | voice-face association model · 1:2 matching and verification, especially gender-restricted queries | accuracy and AUC | Improved |
| Add implicit modality alignment (ablation) | voice-face association model · 1:2 matching | accuracy | Improved |
| Add identity reweighting (ablation) | voice-face association model · 1:2 matching and verification | accuracy and AUC: +1% | Improved |
| Add identity reweighting (ablation) | voice-face association model · retrieval | mAP | Hurt |
| Replace identity reweighting with focal loss (ablation) | voice-face association model · 1:2 matching, verification and retrieval | performance | No clear effect |
| Weight difficult identities without excluding extremely hard identities (R_keep 1.0, reweighting enabled) (ablation) | voice-face association model · 1:2 matching V-F (U) | accuracy: 85.5 -> 86.7 | Improved |
| Weight difficult identities without excluding extremely hard identities (R_keep 1.0, reweighting enabled) (ablation) | voice-face association model · 1:2 matching F-V (U) | accuracy: 85.2 -> 86.8 | Improved |
| Filter identities to retain 0.9 of them (ablation) | identity-reweighted voice-face model · 1:2 matching V-F (U) | accuracy: 86.7 -> 87.2 | Improved |
| Filter identities to retain 0.9 of them (ablation) | identity-reweighted voice-face model · 1:2 matching F-V (U) | accuracy: 86.8 -> 87.2 | Improved |
| Reduce retained identity ratio from 0.9 to 0.6 (ablation) | identity-reweighted voice-face model · 1:2 matching | accuracy: 87.2 -> 85.9; 87.2 -> 85.3 | Hurt |
| Increase retained identity ratio from 0.9 to 1.0 (ablation) | identity-reweighted voice-face model · 1:2 matching | accuracy: 87.2 -> 86.7; 87.2 -> 86.8 | Mixed |
Closest papers by meaning
- Seeing Voices and Hearing Faces: Cross-Modal Biometric Matching (CVPR 2018 · similarity 0.90)
- Seeing Your Speech Style: A Novel Zero-Shot Identity-Disentanglement Face-based Voice Conversion (AAAI 2025 · similarity 0.86)
- FaceSpeak: Expressive and High-Quality Speech Synthesis from Human Portraits of Different Styles (AAAI 2025 · similarity 0.85)
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.84)
- Emotional Face-to-Speech (ICML 2025 · similarity 0.84)
- Identifiable Shared Component Analysis of Unpaired Multimodal Mixtures (NeurIPS 2024 · 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.
