Skip to main content

Advertisement

Table 8 WERs (%) of ESniff MMSE

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

Framework WER (%)
  Set A Set B Set C Average
MTR 7.70 8.23 9.26 8.22
ESniff MMSR (Conventional) 7.59 10.66 8.57 9.01
ESniff SNR-MMSR (SNR Mapping) 6.78 9.56 8.17 8.17
ESniff MMSE (M = 128) 6.71 8.98 7.92 7.86
ESniff MMSE (M = 32) 6.69 9.01 8.00 7.88
ESniff MMSE (M = 16) 6.68 9.02 8.02 7.88
ESniff MMSE (M = 8) 6.61 8.97 8.01 7.86
ESniff MMSE (M = 4) 6.63 9.02 8.03 7.86
ESniff MMSE (M = 2) 8.01 10.58 9.56 9.34
  1. The number of Gaussian distributions varies with MMSE performed along with a comparison to MTR, conventional ESniff MMSR and ESniff SNR-MMSR when using the AFE feature vectors.