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

EURASIP Journal on Audio, Speech, and Music Processing

Table 10 The recognition accuracy results (%) of various forms of WS-HEQ at different SNRs

Method		Clean	20 dB	15 dB	10 dB	5 dB	0 dB	−5 dB
${WS-HEQ}_{I}^{(1)}$	α=0.6	98.99	97.60	95.72	91.54	80.62	55.87	23.26
	Optimal α	99.08	97.76	95.91	91.78	81.09	56.82	24.29
${WS-HEQ}_{I}^{(2)}$	α=0.6	99.09	97.68	95.64	91.30	79.23	49.97	17.69
	Optimal α	99.10	97.86	95.94	91.66	79.87	52.22	19.62
${WS-HEQ}_{I}^{(3)}$	α=0.5	98.99	97.51	95.71	91.59	80.59	55.95	23.88
	Optimal α	99.08	97.67	95.88	91.81	81.03	56.69	24.64
${WS-HEQ}_{I}^{(4)}$	α=0.7	99.07	97.87	96.18	91.98	79.99	51.94	19.83
	Optimal α	99.10	97.87	96.23	92.14	80.51	53.86	23.00
${WS-HEQ}_{II}^{(1)}$	α=0.6	99.05	97.53	95.75	91.99	81.35	58.35	26.98
	Optimal α	99.15	97.62	95.90	92.14	81.65	58.99	28.12
${WS-HEQ}_{II}^{(2)}$	α=0.6	99.04	97.60	95.70	91.68	80.83	56.61	24.45
	Optimal α	99.13	97.80	96.09	92.15	81.44	56.68	24.40
${WS-HEQ}_{II}^{(3)}$	α=0.7	99.05	97.87	95.98	91.86	80.62	53.88	20.51
	Optimal α	99.13	97.51	95.68	91.66	80.94	57.64	27.09
${WS-HEQ}_{II}^{(4)}$	α=0.6	98.96	97.49	95.54	91.17	79.00	51.23	19.37
	Optimal α	99.08	97.58	95.69	91.30	79.41	53.39	21.84

These results are obtained using (1) the scaling factor α listed in Table 1 (2) the scaling factor α that achieves the optimal recognition accuracy with respect to the individual noise type and level (SNR), both of which are for different SNR conditions while averaged over ten noise types as to the Aurora-2 database.