Skip to main content

Advertisement

Table 4 WER (%) of conventional ESniff MMSR method for Aurora 2 task using FE feature vectors

From: Compensation of SNR and noise type mismatch using an environmental sniffing based speech recognition solution

SNR (dB) Set A Set B Set C
  Noise type Noise type Noise type
  Sub. Bab. Car Exh. Ave. Res. Str. Air. Sta. Ave. Sub. Str. Ave.
20 1.69 1.81 1.58 2.04 1.78 1.93 2.54 2.42 2.56 2.36 2.09 2.75 2.42
15 2.43 2.27 1.91 2.72 2.33 4.70 3.87 4.03 4.66 4.31 3.59 4.75 4.17
10 4.08 4.44 3.52 4.94 4.24 9.52 8.13 7.46 8.98 8.52 5.93 9.79 7.86
5 7.58 12.76 7.99 9.60 9.48 24.69 20.62 17.89 18.98 20.54 13.48 24.26 18.97
0 22.63 41.48 20.19 22.74 26.76 54.90 49.82 42.35 42.83 47.47 30.06 54.02 42.04
Ave. 7.68 12.55 7.04 8.41 8.92 19.15 17.00 14.83 15.60 16.64 11.03 19.15 15.09