Research map / Vision-language and multimodal / Cross-modal learning
Unpaired Image-to-Speech Synthesis With Multimodal Information Bottleneck
Shuang Ma, Daniel McDuff, Yale Song
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 the memory fusion module M (ablation) | Ours · image-to-text (COCO) | BLEU-1: 65.2 -> 74.1 | Improved |
| add the memory fusion module M (ablation) | Ours · speech-to-text | WER: 6.99 -> 3.88 | Improved |
| use memory module M instead of two FC layers of similar size (ablation) | Ours · image-to-text | BLEU-1: 65.9 -> 74.1 | Improved |
| add the modality transformer T (ablation) | Ours · image-to-text | BLEU-1: 68.6 -> 74.1 | Improved |
| add the modality transformer T (ablation) | Ours · speech-to-text | WER: 6.01 -> 3.88 | Improved |
| add the adversarial loss L_adv to train T (ablation) | Ours · image-to-text | BLEU-1: 69.8 -> 74.1 | Improved |
| add the adversarial loss L_adv to train T (ablation) | Ours · speech-to-text | WER: 5.87 -> 3.88 | Improved |
| use end-to-end skip-modal model instead of piecewise image-to-text then text-to-speech | image-to-speech · COCO + EMT-4 | WER: 4.1 -> 3.9 | Improved |
| use skip-modal model instead of piecewise baseline | image-to-speech · COCO + EMT-4 | BLEU-4: 30.1 -> 29.9 | No clear effect |
| use CycleGAN cross-domain translation for image-to-speech | image-to-speech · COCO + EMT-4 | BLEU-1: 69.2 -> 26.2 | Hurt |
| swap the 3-layer CNN encoder for UD encoder | Ours · COCO image captioning | BLEU-4: 30.6 -> 37.1 | Improved |
| mix both datasets in each batch instead of alternating (ablation) | Ours · speech-to-text | WER: 3.88 -> 3.76 | Improved |
Closest papers by meaning
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.88)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.88)
- Towards General Modality Translation with Contrastive and Predictive Latent Diffusion Bridge (NeurIPS 2025 · similarity 0.88)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.88)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.87)
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model (CVPR 2025 · similarity 0.87)
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.
