Research map / Image generation / Autoregressive and latent generation
SLD-L2S: Hierarchical Subspace Latent Diffusion for High-Fidelity Lip to Speech Synthesis
Yifan Liang, Andong Li, Kang Yang, Guochen Yu, Fangkun Liu, Lingling Dai, Xiaodong Li, Chengshi Zheng
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 DiCB backbone instead of parameter-matched DiT (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.0835 -> 4.2096 | Improved |
| use DiCB backbone instead of DiT (ablation) | SLD-L2S · LRS3-TED | SECS: 0.618 -> 0.804 | Improved |
| use reparameterized flow matching instead of velocity prediction (ablation) | SLD-L2S · LRS3-TED | UTMOS: 3.9540 -> 4.2096 | Improved |
| use reparameterized flow matching instead of velocity prediction (ablation) | SLD-L2S · LRS3-TED | WER: 29.70 -> 30.22 | No clear effect |
| add SLM loss (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.2885 -> 4.2096 | Hurt |
| add SLM loss (ablation) | SLD-L2S · LRS3-TED | D-BERT and SECS: D-BERT 0.8207 -> 0.8284; SECS 0.776 -> 0.804 | Improved |
| add semantic loss (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.1859 -> 4.2096 | Mixed |
| use 8 subspaces instead of 32 (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.1288 -> 4.2096 | Improved |
| use 8 subspaces instead of 4 (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.1797 -> 4.2096 | Improved |
| vary DiCB kernel size (3x5 to 9x11) (ablation) | SLD-L2S · LRS3-TED | UTMOS: 4.2120 / 4.1914 / 4.2008 vs 4.2096 | No clear effect |
| SLD-L2S vs V2SFlow in subjective intelligibility | SLD-L2S · LRS3 MOS | intelligibility MOS: 3.69 (V2SFLow) vs 3.65 | Hurt |
Closest papers by meaning
- Let There Be Sound: Reconstructing High Quality Speech from Silent Videos (AAAI 2024 · similarity 0.87)
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.86)
- Efficient Speech Language Modeling via Energy Distance in Continuous Latent Space (NeurIPS 2025 · similarity 0.86)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.86)
- ProsodyTalker: 3D Visual Speech Animation via Prosody Decomposition (AAAI 2025 · similarity 0.86)
- Learning Visual Speech Representations via Cross-Modal Distillation and Joint Face-Lip Modeling (NeurIPS 2026 · 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.
