Research map / Architectures and efficiency / Pruning and efficient tuning
Bridging the Gap between Pre-Training and Fine-Tuning for End-to-End Speech Translation
Chengyi Wang, Yu Wu, Shujie Liu, Zhenglu Yang, Ming Zhou
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 TCEN instead of pretrain+MTL | LSTM speech translation · ST-TED En-De (subword average) | BLEU: +3.6 BLEU | Improved |
| Use TCEN instead of pretrain+MTL | LSTM speech translation with character decoding · ST-TED En-De (character average) | BLEU: +2.2 BLEU | Improved |
| Use TCEN instead of pretrain+MTL | Transformer speech translation · ST-TED En-De tst2013 | BLEU: 17.11 vs 14.98; 2.1 BLEU | Improved |
| Use TCEN instead of the cascaded ASR-MT system | speech translation · ST-TED En-De | BLEU: our model outperforms the simple cascaded model over 2 BLEU scores | Improved |
| Use TCEN instead of cascaded+re-seg | speech translation · ST-TED En-De tst2010, tst2013, tst2014, tst2015 | BLEU: 17.61 vs 17.12; 17.67 vs 17.77; 15.73 vs 14.94; 14.94 vs 15.01 | Mixed |
| Use TCEN instead of the strongest listed prior E2E result | LSTM speech translation · Librispeech En-Fr test | BLEU: 17.05 vs 17.02 | No clear effect |
| Add noisy CTC-path MT inputs (ablation) | TCEN · ST-TED En-De tst2013 | BLEU: 14.95 -> 15.50 | Improved |
| Share CTC classification and source embedding weights (ablation) | TCEN · ST-TED En-De tst2013 | BLEU: 14.02 -> 15.50 | Improved |
| Pre-train the model before fine-tuning (ablation) | TCEN · ST-TED En-De tst2013 | BLEU: 8.42 -> 15.50 | Improved |
Closest papers by meaning
- Pre-training for Speech Translation: CTC Meets Optimal Transport (ICML 2023 · similarity 0.89)
- Revisiting End-to-End Speech-to-Text Translation From Scratch (ICML 2022 · similarity 0.88)
- High-Fidelity Simultaneous Speech-To-Speech Translation (ICML 2025 · similarity 0.86)
- Divergence-Guided Simultaneous Speech Translation (AAAI 2024 · similarity 0.85)
- TransVIP: Speech to Speech Translation System with Voice and Isochrony Preservation (NeurIPS 2024 · similarity 0.85)
- Joint Fine-tuning and Conversion of Pretrained Speech and Language Models towards Linear Complexity (ICLR 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.
