Research map / Vision-language and multimodal / Cross-modal learning
Seeing Voices and Hearing Faces: Cross-Modal Biometric Matching
Arsha Nagrani, Samuel Albanie, Andrew Zisserman
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 dynamic face videos instead of still face images | Cross-modal CNN (V-F) · Cross-modal matching, total test set | Identification accuracy: 81.0 -> 84.3 | Improved |
| use dynamic face videos instead of still face images | Cross-modal CNN (F-V) · Cross-modal matching, total test set | Identification accuracy: 79.5 -> 82.9 | Improved |
| remove gender, age, and nationality variation from test pairs | Cross-modal CNN (V-F) · Static test | Identification accuracy: 81.0 -> 63.9 | Hurt |
| remove gender, age, and nationality variation from test pairs | Cross-modal CNN (V-F) · Dynamic test | Identification accuracy: 84.3 -> 67.4 | Hurt |
| use one dynamic image instead of RGB frames (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 79.2 ± 0.1 -> 76.9 ± 0.6 | Hurt |
| use multiple dynamic images instead of one per face-track (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 76.9 ± 0.6 -> 79.9 ± 0.2 | Improved |
| add multiple dynamic images to RGB frame inputs (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 79.2 ± 0.1 -> 84.3 ± 0.2 | Improved |
| add one dynamic image to RGB frame inputs (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 79.2 ± 0.1 -> 82.4 ± 0.3 | Improved |
| replace one dynamic image with multiple dynamic images in RGB fusion (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 82.4 ± 0.3 -> 84.3 ± 0.2 | Improved |
| use dynamic images without RGB face inputs (ablation) | Dynamic matching F-V · Dynamic test | Identification accuracy: 79.9 ± 0.2 | Improved |
Closest papers by meaning
- Seeking the Shape of Sound: An Adaptive Framework for Learning Voice-Face Association (CVPR 2021 · similarity 0.90)
- Speech2Face: Learning the Face Behind a Voice (CVPR 2019 · similarity 0.90)
- Cross-Modal Perceptionist: Can Face Geometry Be Gleaned From Voices? (CVPR 2022 · similarity 0.89)
- Disjoint Mapping Network for Cross-modal Matching of Voices and Faces (ICLR 2019 · similarity 0.89)
- FaceSpeak: Expressive and High-Quality Speech Synthesis from Human Portraits of Different Styles (AAAI 2025 · similarity 0.85)
- Faces that Speak: Jointly Synthesising Talking Face and Speech from Text (CVPR 2024 · 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.
