Table 2 Recognition accuracy (%) achieved by various approaches for Aurora-2 clean-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 | 59.24 | 56.37 | 67.53 | 59.75 | – |
| Â | Spectral-domain methods | Â | Â | ||
SSBoll | 61.81 | 64.09 | 60.09 | 62.38 | 6.53 |
SSBerouti | 69.76 | 70.47 | 69.39 | 69.97 | 25.40 |
SSKamath | 66.91 | 67.50 | 67.19 | 67.20 | 18.52 |
WFPSNR | 71.78 | 73.66 | 70.37 | 72.25 | 31.05 |
WFTSNR | 51.12 | 55.90 | 45.64 | 51.94 | -19.41 |
WFHRNR | 56.20 | 59.65 | 53.47 | 57.03 | -6.74 |
MMSE log-STSA | 72.71 | 73.58 | 71.99 | 72.91 | 32.71 |
MSE | 77.76 | 79.89 | 69.42 | 76.94 | 42.72 |
| Â | Cepstral-domain methods | Â | Â | ||
MVN | 73.81 | 75.02 | 75.08 | 74.55 | 36.77 |
HEQ | 81.42 | 83.34 | 81.51 | 82.21 | 55.80 |
MVA | 78.15 | 79.17 | 79.12 | 78.75 | 47.21 |