AI Paper Atlas

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

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Sign and cross-lingual translation · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
add MT task to ST-only training (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 26.48 -> 26.31No clear effect
add dual-decoder-style DDM distillation (DDM) (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 26.31 -> 27.87Improved
add ASR task (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 27.87 -> 28.29Improved
freeze speech encoder in first epochs (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 28.29 -> 28.64Improved
add MT output regularization with frozen fine-tuned mBART-50 (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 28.64 -> 29.40Improved
add cross-modality learning (CML) loss (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 29.40 -> 29.69Improved
add pseudo ST data from Common Voice (ablation)ComSL Medium · CoVoST 2 ComSL MediumST BLEU: 29.69 -> 30.77Improved
CML vs token-level aligned MMLComSL Medium · CoVoST 2 ComSL MediumAvg BLEU: 29.60 -> 29.69Improved
contrastive ConST instead of CMLComSL Medium · CoVoST 2 ComSL MediumAvg BLEU: 29.40 -> 29.18Hurt
contrastive WACO instead of CMLComSL Medium · CoVoST 2 ComSL MediumAvg BLEU: 29.40 -> 29.37No clear effect
ComSL Large vs Whisper Large fine-tuned, with pseudo dataComSL Large · CoVoST 2 xx-enAvg BLEU: 29.7 -> 31.5Improved
ComSL Large vs USM with fewer parametersComSL Large · CoVoST 2 xx-enAvg BLEU: 30.7 -> 31.5Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy