Automatic Speech Recognition on LibriSpeech Psychoacoustic attack
0WERCTC-FP32
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CTC-FP32Inference precision=FP322026.03 | 0 | 0 | 19.94 | |
| CTC-FP16Inference precision=FP162026.03 | 0 | 0 | 19.37 | |
| CTC-BF16Inference precision=BF162026.03 | 0 | 0 | 18.79 | |
| seq2seq-FP32Inference precision=FP322026.03 | 0 | 0 | 13.85 | |
| seq2seq-BF16Inference precision=BF162026.03 | 0 | 0 | 17.81 | |
| Transformer-FP32Inference precision=FP322026.03 | 0 | 0 | 24.84 | |
| Transformer-FP16Inference precision=FP162026.03 | 0 | 0 | 24.23 | |
| Transformer-BF16Inference precision=BF162026.03 | 0 | 0 | 25.04 | |
| Whisper-FP32Inference precision=FP322026.03 | 0 | 0 | 23.88 | |
| Whisper-BF16Inference precision=BF162026.03 | 0 | 0 | 23.72 | |
| seq2seq-FP16Inference precision=FP162026.03 | 0.21 | 1 | 16.33 | |
| Transformer-FP32Inference precision=FP162026.03 | 6.58 | 7 | 24.84 | |
| Transformer-FP16Inference precision=FP322026.03 | 7.41 | 9 | 24.23 | |
| PVP (Transformer-FP16)Inference precision=Random2026.03 | 7.49 | 12.4 | 24.23 | |
| CTC-FP32Inference precision=FP162026.03 | 8.02 | 36 | 19.94 | |
| CTC-FP16Inference precision=FP322026.03 | 9.47 | 39 | 19.37 | |
| PVP (Transformer-BF16)Inference precision=Random2026.03 | 10.31 | 14.9 | 25.04 | |
| Whisper-FP16Inference precision=FP162026.03 | 10.36 | 8 | 33.23 | |
| PVP (CTC-FP32)Inference precision=Random2026.03 | 10.74 | 39.3 | 19.94 | |
| PVP (Transformer-FP32)Inference precision=Random2026.03 | 11.15 | 13 | 24.84 | |
| PVP (CTC-FP16)Inference precision=Random2026.03 | 13.81 | 44 | 19.37 | |
| PVP (CTC-BF16)Inference precision=Random2026.03 | 13.87 | 53.3 | 18.79 | |
| Transformer-FP16Inference precision=BF162026.03 | 15.23 | 26 | 24.23 | |
| Transformer-BF16Inference precision=FP322026.03 | 16.46 | 25 | 25.04 | |
| Transformer-BF16Inference precision=FP162026.03 | 16.46 | 25 | 25.04 | |
| CTC-BF16Inference precision=FP322026.03 | 20.78 | 81 | 18.79 | |
| CTC-BF16Inference precision=FP162026.03 | 21.4 | 83 | 18.79 | |
| PVP (seq2seq-FP32)Inference precision=Random2026.03 | 21.52 | 47.8 | 13.85 | |
| Whisper-BF16Inference precision=FP322026.03 | 22.31 | 21 | 23.72 | |
| Transformer-FP32Inference precision=BF162026.03 | 24.28 | 28.8 | 24.84 | |
| CTC-FP32Inference precision=BF162026.03 | 25.31 | 84 | 19.94 | |
| PVP (Whisper-FP32)Inference precision=Random2026.03 | 26.89 | 23 | 23.88 | |
| PVP (Whisper-BF16)Inference precision=Random2026.03 | 26.97 | 24.9 | 23.72 | |
| seq2seq-FP32Inference precision=FP162026.03 | 27.16 | 58 | 13.85 | |
| CTC-FP16Inference precision=BF162026.03 | 30.25 | 89 | 19.37 | |
| PVP (seq2seq-BF16)Inference precision=Random2026.03 | 33.15 | 63.3 | 17.81 | |
| Whisper-FP32Inference precision=BF162026.03 | 35.06 | 29 | 23.88 | |
| seq2seq-FP32Inference precision=BF162026.03 | 35.19 | 79 | 13.85 | |
| PVP (seq2seq-FP16)Inference precision=Random2026.03 | 40.21 | 65.9 | 16.33 | |
| PVP (Whisper-FP16)Inference precision=Random2026.03 | 44.16 | 45.8 | 33.23 | |
| seq2seq-BF16Inference precision=FP162026.03 | 50.41 | 97 | 17.81 | |
| seq2seq-BF16Inference precision=FP322026.03 | 50.62 | 97 | 17.81 | |
| seq2seq-FP16Inference precision=FP322026.03 | 57.41 | 98 | 16.33 | |
| Whisper-FP16Inference precision=FP322026.03 | 58.37 | 60 | 33.23 | |
| Whisper-BF16Inference precision=FP162026.03 | 59.36 | 53 | 23.72 | |
| Whisper-FP16Inference precision=BF162026.03 | 59.76 | 65 | 33.23 | |
| seq2seq-FP16Inference precision=BF162026.03 | 60.7 | 93 | 16.33 | |
| Whisper-FP32Inference precision=FP162026.03 | 60.76 | 52 | 23.88 |