Research map / Large language models / Language, safety and interpretability
Confident and Adaptive Generative Speech Recognition via Conformal Risk Control
Amit Damri, Bracha Laufer-Goldshtein
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 LTT adaptive hypothesis selection instead of fixed N=5 | GER LLaMA-2-7B · TedLium-3 | average set size: 5 -> 2.48 (-50.08%) | Improved |
| add LTT adaptive selection instead of fixed N=5 | GER LLaMA-2-7B · TedLium-3 | WER (%): 7.53 -> 7.52 | No clear effect |
| add LTT adaptive selection instead of fixed N=5 | GER LLaMA-2-7B · CHiME-4 | WER (%): 6.24 -> 6.37 | Hurt |
| add LTT adaptive selection instead of fixed N=5 | GER LLaMA-2-7B · CommonVoice | WER (%): 8.32 -> 8.51 | Hurt |
| add LTT adaptive selection instead of fixed N=5 | GER LLaMA-2-7B · CommonVoice | average set size: 5 -> 3.29 (-34.2%) | Improved |
| raise risk tolerance alpha from 0.21 to 0.24 (ablation) | LTT adaptive selection · CommonVoice | WER (%): set size 3.73 -> 2.99; WER 8.53 -> 8.73 | Mixed |
| train on constant-5 hypothesis sets instead of dynamic sizes (ablation) | GER LLaMA-2-7B · CHiME-4 | avg WER (%): 8.43 -> 7.79 | Improved |
| train on 5 hypotheses instead of 1 hypothesis (ablation) | GER LLaMA-2-7B · CHiME-4 (5-hyp test) | WER (%): 11.65 -> 6.38 | Improved |
| train and test with matched size 3 instead of train-5 (ablation) | GER LLaMA-2-7B · CHiME-4 (3-hyp test) | WER (%): 6.74 -> 6.58 | No clear effect |
| scale GER model from 7B to 13B | LTT adaptive selection · TedLium-3 | WER (%): 7.52 -> 6.46 | Improved |
| apply adaptive selection to GPT-3.5-turbo zero-shot | LTT adaptive selection · CommonVoice | WER (%): 11.73 -> 11.81 | No clear effect |
| apply adaptive selection to speech translation (TER risk) | GenTranslate · FLEURS ar->en | BLEU: 34.47 -> 33.44 | Hurt |
Closest papers by meaning
- What Was That Again? Certified Robustness for Automatic Speech Recognition (NeurIPS 2026 · similarity 0.88)
- Large Language Models are Efficient Learners of Noise-Robust Speech Recognition (ICLR 2024 · similarity 0.88)
- It's Never Too Late: Fusing Acoustic Information into Large Language Models for Automatic Speech Recognition (ICLR 2024 · similarity 0.88)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.86)
- Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models (NeurIPS 2024 · similarity 0.86)
- Speech Robust Bench: A Robustness Benchmark For Speech Recognition (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.
