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

Table 2 Word accuracy rates of the Aurora 2.0 clean-train tasks for the 0-20 dB SNR test data, using the proposed method

From: Noise-robust speech feature processing with empirical mode decomposition

 

Set A

Set B

Set C

Avg.

Rel. imp.

Baseline

61.3

55.8

66.1

60.1

=

MVN

70.2

70.8

66.4

69.7

24.0

MVN+EMD(e)

76.8

76.7

75.3

76.5

41.1

AFE

87.5

87.0

85.6

86.9

67.1

AFE+EMD

87.6

86.6

86.1

86.9

67.1

  1. Baseline: raw features; MVN: mean variance normalization; MVN+EMD(e): EMD applied on the log-energy sequence; AFE: advanced frontend; AFE+EMD: EMD applied on the combined C0/log-energy sequence of AFE.
Back to article page