Research map / Architectures and efficiency / RNNs and normalization
Regularizing End-to-End Speech Translation with Triangular Decomposition Agreement
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 triangular-decomposition KL agreement regularization (ablation) | E2E-ST medium · MuST-C tst-COMMON EN-DE | BLEU: 23.8 -> 25.4 | Helped |
| Add triangular-decomposition KL agreement regularization (ablation) | E2E-ST medium · MuST-C tst-COMMON EN-FR | BLEU: 34.5 -> 36.1 | Helped |
| Add triangular-decomposition KL agreement regularization (ablation) | E2E-ST medium · MuST-C tst-COMMON EN-DE | WER: 16.3 -> 14.9 | Helped |
| Add triangular-decomposition KL agreement regularization (ablation) | E2E-ST medium · MuST-C tst-COMMON EN-FR | WER: 16.0 -> 14.1 | Helped |
| Replace the baseline training process with joint ASR-ST training | E2E-ST small · MuST-C tst-COMMON average | BLEU: 24.8 -> 25.1 | Helped |
| Replace the baseline training process with joint ASR-ST training | E2E-ST medium · MuST-C tst-COMMON average | BLEU: 25.4 -> 25.3 | Hurt |
| Add word-level knowledge distillation (ablation) | E2E-ST-Base medium · MuST-C tst-COMMON average EN-DE/EN-FR | BLEU: 28.7 -> 29.2 | Helped |
| Add sequence-level knowledge distillation (ablation) | E2E-ST-Base medium · MuST-C tst-COMMON average EN-DE/EN-FR | BLEU: 28.7 -> 29.9 | Helped |
| Increase embedding dimension (ablation) | E2E-ST-TDA · MuST-C EN-DE | BLEU | Mixed |
| Vary the KL-loss weight; λ = 1.0 is best among tested settings (ablation) | E2E-ST-TDA · MuST-C EN-DE development set | BLEU | Mixed |
| Use TDA instead of the baseline method | E2E-ST medium · MuST-C tst-COMMON EN-DE/EN-FR extended setting | BLEU: 25.8 -> 27.1; 35.9 -> 37.4 | Helped |
Closest papers by meaning
- Revisiting End-to-End Speech-to-Text Translation From Scratch (ICML 2022 · similarity 0.91)
- Divergence-Guided Simultaneous Speech Translation (AAAI 2024 · similarity 0.88)
- High-Fidelity Simultaneous Speech-To-Speech Translation (ICML 2025 · similarity 0.86)
- DiffNorm: Self-Supervised Normalization for Non-autoregressive Speech-to-speech Translation (NeurIPS 2024 · similarity 0.86)
- TransVIP: Speech to Speech Translation System with Voice and Isochrony Preservation (NeurIPS 2024 · similarity 0.86)
- FastLongSpeech: Enhancing Large Speech-Language Models for Efficient Long-Speech Processing (NeurIPS 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.