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

EURASIP Journal on Audio, Speech, and Music Processing

Table 12 Recognition accuracy results (%) of WS-HEQ for different SNR conditions at white noise environment

Method		Clean	20 dB	15 dB	10 dB	5 dB	Average
MFCC		76.38	30.08	14.72	6.23	3.52	13.64
CHN		76.52	55.22	43.56	30.90	18.84	37.13
${WS-HEQ}_{I}^{(1)}$	α=1.0	77.15	56.48	46.81	33.05	19.40	38.94
	α=0.6	76.94	59.28	49.81	36.80	22.74	42.16
${WS-HEQ}_{I}^{(2)}$	α=1.0	77.50	56.30	47.62	33.75	19.82	39.37
	α=0.6	76.07	59.63	49.88	36.73	23.74	42.50
${WS-HEQ}_{I}^{(3)}$	α=1.0	77.08	55.92	44.94	32.09	19.36	38.08
	α=0.6	76.84	58.68	48.44	36.22	23.11	41.61
${WS-HEQ}_{I}^{(4)}$	α=1.0	76.91	54.71	45.29	32.02	19.22	37.81
	α=0.6	76.59	58.14	48.95	35.89	23.11	41.52
${WS-HEQ}_{II}^{(1)}$	α=1.0	77.96	56.88	47.46	33.82	20.94	39.78
	α=0.6	76.54	59.75	49.79	36.89	23.72	42.54
${WS-HEQ}_{II}^{(2)}$	α=1.0	77.31	57.79	47.83	33.91	20.85	40.10
	α=0.6	76.33	59.86	49.72	37.24	24.37	42.80
${WS-HEQ}_{II}^{(3)}$	α=1.0	77.92	57.18	46.36	33.70	20.64	39.47
	α=0.6	77.01	59.98	49.11	36.26	23.53	42.22
${WS-HEQ}_{II}^{(4)}$	α=1.0	77.10	57.11	46.46	34.40	20.90	39.72
	α=0.6	77.24	60.07	49.91	36.85	23.65	42.62

These recognition accuracy results (%) of the MFCC baseline, CHN, and eight forms of WS-HEQ are for different SNR conditions at the white noise environment as to the subset of the TCC database.