Automatic Speech Recognition on LibriSpeech-FreeSound (car)
2.4WER (%)Whisper (real label)
Evaluation results
| Method | Links | |
|---|---|---|
| Whisper (real label)Adaptation strategy=Finetuning, Supervision type=Supervised2024.05 | 2.4 | |
| STARAdaptation strategy=STAR, Supervision type=Unsupervised2024.05 | 2.5 | |
| TOKreweight (A_t)Adaptation strategy=Self-training, Token-level re-weighting=A_t, Supervision type=Unsupervised2024.05 | 2.6 | |
| UTTfilterAdaptation strategy=Self-training, Filtering=Utterance-level, Supervision type=Unsupervised2024.05 | 2.8 | |
| TOKreweight (G_t)Adaptation strategy=Self-training, Token-level re-weighting=G_t, Supervision type=Unsupervised2024.05 | 2.8 | |
| Whisper (frozen)Adaptation strategy=None, Supervision type=Zero-shot2024.05 | 2.9 | |
| Whisper (self-train.)Adaptation strategy=Self-training, Supervision type=Unsupervised2024.05 | 3 |