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

EURASIP Journal on Audio, Speech, and Music Processing

Table 5 The recognition accuracy results (%) achieved by the combination of MVA and WS-HEQ

Method	Set A	Set B	Set C	Average
MVA	78.15	79.17	79.12	78.75
${WS-HEQ}_{I}^{(1)}$	83.36	85.37	83.89	84.27
${WS-HEQ}_{I}^{(1)}$ +MVA	84.95	85.60	84.97	85.21
${WS-HEQ}_{I}^{(2)}$	82.29	83.22	82.82	82.76
${WS-HEQ}_{I}^{(2)}$ +MVA	85.34	85.89	85.69	85.63
${WS-HEQ}_{I}^{(3)}$	83.57	85.15	83.93	84.27
${WS-HEQ}_{I}^{(3)}$ +MVA	84.84	85.59	85.38	85.25
${WS-HEQ}_{I}^{(4)}$	82.88	84.70	82.78	83.59
${WS-HEQ}_{I}^{(4)}$ +MVA	84.62	85.70	84.90	85.11
${WS-HEQ}_{II}^{(1)}$	84.13	86.16	84.39	84.99
${WS-HEQ}_{II}^{(1)}$ +MVA	85.08	86.43	85.40	85.69
${WS-HEQ}_{II}^{(2)}$	83.54	85.75	83.83	84.48
${WS-HEQ}_{II}^{(2)}$ +MVA	85.81	86.69	86.17	86.23
${WS-HEQ}_{II}^{(3)}$	83.25	85.10	83.50	84.04
${WS-HEQ}_{II}^{(3)}$ +MVA	84.18	85.69	84.48	84.84
${WS-HEQ}_{II}^{(4)}$	82.30	83.45	82.90	82.88
${WS-HEQ}_{II}^{(4)}$ +MVA	85.14	85.41	85.87	85.40

They are for different test sets while averaged over five SNR conditions (20 to 0 dB) as to the Aurora-2 database. The scaling factor α listed in Table 1 is adopted for each WS-HEQ.