Research map / Label-efficient and robust learning / Recognition and long-tail
Robust Speech Recognition via Large-Scale Weak Supervision
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 |
|---|---|---|---|
| Train on broad, diverse weakly supervised audio and evaluate zero-shot | Whisper · 12 out-of-distribution English speech datasets | WER: 29.3 -> 12.8 | Helped |
| Use zero-shot multilingual pretraining | Whisper · Multilingual LibriSpeech (MLS) | WER: 9.7 -> 7.3 | Helped |
| Use zero-shot multilingual pretraining | Whisper · VoxPopuli | WER: 8.1 -> 13.6 | Hurt |
| Train with large-scale X→English translation supervision | Whisper · CoVoST2 X→English, all languages | BLEU: 25.2 -> 29.1 | Helped |
| Use zero-shot large-scale training | Whisper · CoVoST2 X→English, high-resource | BLEU: 38.2 -> 36.2 | Hurt |
| Use Whisper's language identification model | Whisper · Fleurs | language identification accuracy: 77.7 -> 64.5 | Hurt |
| Add beam search to long-form decoding (ablation) | Whisper · Long-form transcription, average across seven datasets | WER: 11.0 -> 10.6 | Helped |
| Add temperature fallback after beam search (ablation) | Whisper · Long-form transcription, average across seven datasets | WER: 10.6 -> 10.6 | No clear effect |
| Add voice activity detection (ablation) | Whisper · Long-form transcription, average across seven datasets | WER: 10.6 -> 10.2 | Helped |
| Add previous-text conditioning (ablation) | Whisper · Long-form transcription, average across seven datasets | WER: 10.2 -> 10.0 | Helped |
| Constrain the initial timestamp to between 0.0 and 1.0 second (ablation) | Whisper · Long-form transcription, average across seven datasets | WER: 10.0 -> 10.0 | No clear effect |
| Increase dataset size from 3,405 to 681,070 hours (ablation) | Whisper Medium · English speech recognition, average over 12 datasets | WER: 30.5 -> 9.9 | Helped |
Closest papers by meaning
- Speech Robust Bench: A Robustness Benchmark For Speech Recognition (ICLR 2025 · similarity 0.88)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.87)
- Unified Speech Recognition: A Single Model for Auditory, Visual, and Audiovisual Inputs (NeurIPS 2024 · similarity 0.86)
- The Brain's Bitter Lesson: Scaling Speech Decoding With Self-Supervised Learning (ICML 2025 · similarity 0.86)
- What Was That Again? Certified Robustness for Automatic Speech Recognition (NeurIPS 2026 · similarity 0.86)
- Dynamic-SUPERB Phase-2: A Collaboratively Expanding Benchmark for Measuring the Capabilities of Spoken Language Models with 180 Tasks (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.