Research map / Large language models / Language, safety and interpretability
ComSL: A Composite Speech-Language Model for End-to-End Speech-to-Text Translation
Chenyang Le, Yao Qian, Long Zhou, Shujie LIU, Yanmin Qian, Michael Zeng, Xuedong Huang
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 MT task to ST-only training (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 26.48 -> 26.31 | No clear effect |
| add dual-decoder-style DDM distillation (DDM) (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 26.31 -> 27.87 | Improved |
| add ASR task (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 27.87 -> 28.29 | Improved |
| freeze speech encoder in first epochs (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 28.29 -> 28.64 | Improved |
| add MT output regularization with frozen fine-tuned mBART-50 (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 28.64 -> 29.40 | Improved |
| add cross-modality learning (CML) loss (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 29.40 -> 29.69 | Improved |
| add pseudo ST data from Common Voice (ablation) | ComSL Medium · CoVoST 2 ComSL Medium | ST BLEU: 29.69 -> 30.77 | Improved |
| CML vs token-level aligned MML | ComSL Medium · CoVoST 2 ComSL Medium | Avg BLEU: 29.60 -> 29.69 | Improved |
| contrastive ConST instead of CML | ComSL Medium · CoVoST 2 ComSL Medium | Avg BLEU: 29.40 -> 29.18 | Hurt |
| contrastive WACO instead of CML | ComSL Medium · CoVoST 2 ComSL Medium | Avg BLEU: 29.40 -> 29.37 | No clear effect |
| ComSL Large vs Whisper Large fine-tuned, with pseudo data | ComSL Large · CoVoST 2 xx-en | Avg BLEU: 29.7 -> 31.5 | Improved |
| ComSL Large vs USM with fewer parameters | ComSL Large · CoVoST 2 xx-en | Avg BLEU: 30.7 -> 31.5 | Improved |
Closest papers by meaning
- High-Fidelity Simultaneous Speech-To-Speech Translation (ICML 2025 · similarity 0.88)
- Divergence-Guided Simultaneous Speech Translation (AAAI 2024 · similarity 0.87)
- SimulMEGA: MoE Routers are Advanced Policy Makers for Simultaneous Speech Translation (NeurIPS 2025 · similarity 0.87)
- TransVIP: Speech to Speech Translation System with Voice and Isochrony Preservation (NeurIPS 2024 · similarity 0.87)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.87)
- Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion (ICLR 2026 · similarity 0.86)
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.
