Table 5 Recognition accuracy (%) achieved by various approaches for Aurora-2 multi-condition training task averaged across the SNRs between 0 and 20 dB, where AVG (%) and RR (%) are the averaged accuracy rate and the relative error rate reduction over the baseline
From: Enhancing the magnitude spectrum of speech features for robust speech recognition
Method | Set A | Set B | Set C | AVG | RR |
|---|---|---|---|---|---|
MFCC baseline | 86.10 | 86.05 | 83.88 | 85.64 | – |
SSBerouti | 83.66 | 84.00 | 82.93 | 83.65 | -13.86 |
WFPSNR | 83.96 | 84.50 | 83.03 | 83.99 | -11.49 |
MMSE log-STSA | 81.21 | 82.62 | 80.82 | 81.69 | -27.48 |
MSE | 87.91 | 87.41 | 82.21 | 86.57 | 6.49 |
MVN | 90.38 | 90.41 | 89.82 | 90.28 | 32.31 |
SSBerouti+MVN | 86.89 | 87.91 | 85.72 | 87.06 | 9.91 |
WFPSNR+MVN | 84.78 | 85.24 | 84.57 | 84.92 | -5.00 |
MMSE log-STSA+MVN | 86.99 | 86.82 | 85.99 | 86.72 | 7.53 |
MSE+MVN | 90.00 | 89.59 | 87.01 | 89.24 | 25.07 |
HEQ | 89.98 | 90.05 | 89.59 | 89.93 | 29.87 |
SSBerouti+HEQ | 87.38 | 88.21 | 86.23 | 87.48 | 12.83 |
WFPSNR+HEQ | 84.77 | 85.16 | 84.16 | 84.80 | -5.84 |
MMSE log-STSA+HEQ | 86.42 | 86.53 | 85.27 | 86.24 | 4.15 |
MSE+HEQ | 89.78 | 90.03 | 87.74 | 89.47 | 26.67 |
MVA | 90.97 | 91.04 | 90.85 | 90.98 | 37.19 |
SSBerouti+MVA | 88.14 | 88.69 | 87.37 | 88.21 | 17.90 |
WFPSNR+MVA | 85.80 | 85.68 | 85.38 | 85.67 | 0.20 |
MMSE log-STSA+MVA | 87.40 | 87.17 | 86.59 | 87.14 | 10.46 |
MSE+MVA | 90.69 | 89.75 | 88.28 | 89.83 | 29.17 |