Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 11 The recognition accuracy results (%) of the three schemes defined in Section 5.4

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

 

SNR

The scaling factorα

 
  

0.6

0.8

1.0

1.2

1.4

Scheme 1 (MFCC)

Clean

99.08

99.16

99.12

99.14

99.09

 

20 ∼ 0 dB

52.84

59.15

59.75

59.42

56.48

 

−5 dB

6.13

7.67

8.16

7.70

6.87

Scheme 2 (pre-HEQ)

clean

98.95

99.00

98.97

98.92

99.00

 

20 ∼ 0 dB

81.84

81.41

80.32

80.16

80.04

 

−5 dB

17.58

17.27

19.94

15.12

14.39

Scheme 3 (post-HEQ)

clean

98.97

99.00

98.97

98.99

98.84

 

20 ∼ 0 dB

81.19

80.42

80.32

79.72

79.35

 

−5 dB

16.73

14.62

19.94

14.05

18.34

  1. They are for the clean condition, the average of five SNR conditions (20, 15, 10, 5, and 0 dB), and the −5 dB SNR condition but averaged over the ten noise types as to the Aurora-2 database.
Back to article page