Automatic Speech Recognition on Earnings21 long-form (test)
10.8WERWhisper small.en
Evaluation results
| Method | Links | |
|---|---|---|
| Whisper small.enDecoder=Attention Decoding, Chunk Size (seconds)=302025.12 | 10.8 | |
| Ours.smallDecoder=Joint CTC-Attention, Chunk Size (seconds)=0.962025.12 | 11.4 | |
| Ours.smallDecoder=Joint CTC-Attention, Chunk Size (seconds)=3.842025.12 | 11.4 | |
| Ours.baseDecoder=Joint CTC-Attention, Chunk Size (seconds)=3.842025.12 | 12.1 | |
| Ours.smallDecoder=Attention Decoding, Chunk Size (seconds)=0.962025.12 | 12.1 | |
| Ours.baseDecoder=Joint CTC-Attention, Chunk Size (seconds)=0.962025.12 | 12.2 | |
| Whisper base.enDecoder=Attention Decoding, Chunk Size (seconds)=302025.12 | 12.5 | |
| Ours.smallDecoder=Attention Rescoring, Chunk Size (seconds)=0.962025.12 | 12.5 | |
| Ours.baseDecoder=Attention Decoding, Chunk Size (seconds)=0.962025.12 | 13.3 | |
| Ours.baseDecoder=Attention Rescoring, Chunk Size (seconds)=0.962025.12 | 13.4 |