Speech Recognition with Timestamps on Multilingual LibriSpeech (MLS)
48.36AASCrisperWhisper
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CrisperWhisper2026.04 | 48.36 | — | |
| Mixed TrainingEvaluation Setting=zero-shot2026.04 | 39.72 | 0 | |
| Mixed Training + Length AugLength Augmentation=true, Evaluation Setting=zero-shot2026.04 | 35.13 | 0.03 | |
| Mixed Training + Length Aug + Reduced TFLength Augmentation=true, Reduced Teacher Forcing=true, Evaluation Setting=zero-shot2026.04 | 34.94 | 0.08 | |
| Mixed Training + Length Aug + Timestamp RegLength Augmentation=true, Timestamp Regularization=true, Evaluation Setting=zero-shot2026.04 | 34.72 | 0.03 | |
| Qwen-Audio2026.04 | 31.87 | 4.51 |