Automatic Speech Recognition on MUSAN additive noise on LibriSpeech (dev-other)
6.62Greedy ScoreGreedy decoding
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Greedy decodingG=16, Condition=MUSAN 20 dB, SNR (dB)=20 dB, nbest_scale=1.02026.06 | 6.62 | — | — | — | — | — | |
| Greedy decodingG=16, Condition=MUSAN 10 dB, SNR (dB)=10 dB, nbest_scale=1.02026.06 | 8.04 | — | — | — | — | — | |
| Greedy decodingG=16, Condition=MUSAN 5 dB, SNR (dB)=5 dB, nbest_scale=1.02026.06 | 11.1 | — | — | — | — | — | |
| Greedy decodingG=16, Condition=MUSAN 0 dB, SNR (dB)=0 dB, nbest_scale=1.02026.06 | 17.88 | — | — | — | — | — | |
| MBR-CER + RoBERTa PLL (τ=10)G=16, Condition=MUSAN 20 dB, SNR (dB)=20 dB, nbest_scale=1.02026.06 | — | — | 6.38 | 0.24 | 0.006 | 13.9 | |
| MBR-CER + RoBERTa PLL (τ=10)G=16, Condition=MUSAN 10 dB, SNR (dB)=10 dB, nbest_scale=1.02026.06 | — | — | 7.8 | 0.24 | 0.003 | 13.5 | |
| MBR-CER + RoBERTa PLL (τ=10)G=16, Condition=MUSAN 5 dB, SNR (dB)=5 dB, nbest_scale=1.02026.06 | — | — | 10.84 | 0.27 | 0.001 | 13 | |
| MBR-CER + RoBERTa PLL (τ=10)G=16, Condition=MUSAN 0 dB, SNR (dB)=0 dB, nbest_scale=1.02026.06 | — | — | 17.64 | 0.23 | 0.646 | — | |
| OracleG=16, Condition=MUSAN 20 dB, SNR (dB)=20 dB, nbest_scale=1.02026.06 | — | 4.89 | — | — | — | — | |
| OracleG=16, Condition=MUSAN 10 dB, SNR (dB)=10 dB, nbest_scale=1.02026.06 | — | 6.26 | — | — | — | — | |
| OracleG=16, Condition=MUSAN 5 dB, SNR (dB)=5 dB, nbest_scale=1.02026.06 | — | 9.06 | — | — | — | — | |
| OracleG=16, Condition=MUSAN 0 dB, SNR (dB)=0 dB, nbest_scale=1.02026.06 | — | 15.57 | — | — | — | — |