Skip to main content

Table 8 The recognition accuracy results (%) of WS-HEQ II ( 1 ) using the optimal scaling factor α (in parentheses)

From: Intra-frame cepstral sub-band weighting and histogram equalization for noise-robust speech recognition

SNR

Set A

Set B

Set C

 

Subway

Babble

Car

Exhibition

Restaurant

Street

Airport

Train

MIRS subway

MIRS street

Clean

99.36 (0.7)

99.06 (0.8)

99.05 (0.8)

99.11 (0.4)

99.36 (0.7)

99.06 (0.8)

99.05 (0.8)

99.11 (0.4)

99.17 (0.7)

99.15 (0.4)

20 dB

96.56 (0.7)

97.52 (0.6)

98.12 (0.8)

96.98 (0.8)

97.54 (0.7)

97.94 (0.8)

98.42 (0.5)

98.36 (0.8)

96.96 (0.4)

97.76 (0.8)

15 dB

94.60 (0.4)

95.71 (0.7)

97.17 (0.7)

94.60 (0.5)

95.52 (0.6)

96.55 (0.6)

96.90 (0.6)

97.25 (0.8)

94.41 (0.6)

96.28 (0.6)

10 dB

89.65 (0.5)

92.14 (0.6)

93.83 (0.7)

89.48 (0.5)

91.83 (0.7)

93.05 (0.6)

94.72 (0.8)

94.32 (0.6)

89.75 (0.4)

92.62 (0.6)

5 dB

79.89 (0.5)

80.11 (0.6)

84.28 (0.8)

78.22 (0.5)

80.07 (0.7)

82.56 (0.5)

85.51 (0.8)

84.33 (0.6)

79.83 (0.4)

81.65 (0.6)

0 dB

58.27 (0.4)

53.42 (0.7)

61.65 (0.7)

57.27 (0.4)

56.77 (0.7)

58.92 (0.5)

65.49 (0.7)

61.65 (0.8)

57.72 (0.4)

58.74 (0.6)

−5 dB

28.43 (0.5)

22.76 (0.8)

28.81 (0.9)

30.48 (0.4)

25.85 (0.9)

27.33 (0.6)

31.26 (0.9)

29.28 (0.9)

29.60 (0.5)

27.42 (0.8)

  1. This is with respect to each noise type and level (SNR) as to the Aurora-2 database.