Skip to main content

Table 2 WER of the MC MDT experiments on test set 1-7 (Close-talking microphone)

From: Multi-candidate missing data imputation for robust speech recognition

  WER (%)
  1. Clean 2. Car 3. Babble 4. Restau. 5. Street 6. Airport 7. Train Average
1. No MDT 6.82 12.98 32.62 40.95 38.50 32.51 38.37 28.96 ± 0.5
2. BE PROSPECT MDT 6.33 10.89 20.14 28.38 25.42 20.57 25.14 19.55 ± 0.4
3. GB GD MC MDT 6.28 10.95 20.08 29.76 24.98 20.18 25.28 19.64 ± 0.4
4. GB MAP MC MDT 6.00 10.31 20.90 29.12 24.86 19.73 25.89 19.55 ± 0.4
5. GB GD MC MDT + CGS 6.16 10.93 20.36 29.48 25.72 19.91 25.13 19.53 ± 0.4
6. GD CLBR 6.18 11.96 22.90 31.22 26.60 21.74 29.03 21.38 ± 0.4
7. MAP CLBR 6.28 11.81 21.69 31.44 27.63 21.74 28.51 21.30 ± 0.4
8. SB MC MDT 6.67 11.25 20.25 28.90 24.49 19.67 25.20 19.49 ± 0.4
9. Spectral MC MDT 6.41 10.46 21.63 29.68 25.89 20.66 25.85 20.08 ± 0.4
10. Cepstral MC MDT 6.44 11.04 19.90 29.46 25.61 19.86 26.53 19.83 ± 0.4
  1. GB, Gaussian-based; SB, State-based; CGS, CG selection; CLBR, cluster-based reconstruction; BE, Backend.