Research map / Image generation / Autoregressive and latent generation
From Inference to Generation: End-to-end Fully Self-supervised Generation of Human Face from Speech
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 mismatched-identity loss (relidGANs) during generation training (ablation) | Face generation from speech · AVSpeech face retrieval, Top-1 cosine distance | cosine distance retrieval performance: 7.36 -> 13.59 | Helped |
| Add mismatched-identity loss (relidGANs) during generation training (ablation) | Face generation from speech · AVSpeech test, generated image versus ground-truth image | inference preference (%): 47.2% -> 76.65% | Helped |
| Add mismatched-identity loss (relidGANs) during generation training (ablation) | Face generation from speech · AVSpeech test, comparison of two generated images | inference preference (%): 79.68% | Helped |
| Add mismatched-identity loss (relidGANs) during generation training (ablation) | Face generation from speech · AVSpeech face retrieval, Top-10 cosine distance | cosine distance retrieval performance: 39.59 -> 53.83 | Helped |
| Skip inference-stage training before training the generator (ablation) | Two-stage inference and generation framework · AVSpeech face generation | speech-condition use | Hurt |
| Train using self-supervised cross-modal matching rather than supervised identity labels | Cross-modal identity matching · VoxCeleb ∩ VGGFace, V-F 10-way | accuracy (%): 40 -> 55.66 | Helped |
| Train using self-supervised cross-modal matching rather than supervised identity labels | Cross-modal identity matching · VoxCeleb ∩ VGGFace, V-F 2-way | accuracy (%): 84.12 -> 79.90 | Hurt |
Closest papers by meaning
- Speech2Face: Learning the Face Behind a Voice (CVPR 2019 · similarity 0.90)
- Faces that Speak: Jointly Synthesising Talking Face and Speech from Text (CVPR 2024 · similarity 0.87)
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.86)
- FaceSpeak: Expressive and High-Quality Speech Synthesis from Human Portraits of Different Styles (AAAI 2025 · similarity 0.86)
- AE-NeRF: Audio Enhanced Neural Radiance Field for Few Shot Talking Head Synthesis (AAAI 2024 · similarity 0.85)
- Emotional Face-to-Speech (ICML 2025 · 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.