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

EURASIP Journal on Audio, Speech, and Music Processing

Table 2 The recognition accuracy results (%) of the MFCC baseline, CHN, S-HEQ, and WS-HEQ with structure I

Method		Set A	Set B	Set C	Average	RR
MFCC		59.24	56.37	67.53	59.75	-
CHN		79.28	81.53	79.98	80.32	51.11
${WS-HEQ}_{I}^{(1)}$	α=1.0 (S-HEQ)	81.56	84.51	80.78	82.58	56.73
	α=0.6	83.36	85.37	83.89	84.27	60.92
${WS-HEQ}_{I}^{(2)}$	α=1.0	80.88	83.64	80.46	81.90	55.04
	α=0.6	82.29	83.22	82.82	82.76	57.16
${WS-HEQ}_{I}^{(3)}$	α=1.0	79.66	82.51	79.33	80.73	52.13
	α=0.5	83.57	85.15	83.93	84.27	60.92
${WS-HEQ}_{I}^{(4)}$	α=1.0	80.20	82.82	80.18	81.24	53.39
	α=0.7	82.88	84.70	82.78	83.59	59.23

They are for different test sets while averaged over five SNR conditions (20 to 0 dB) as to the Aurora-2 database. RR (%) is the relative error rate reduction compared with the MFCC baseline.