Autocorrelation-based noise subtraction method with smoothing, overestimation, energy, and cepstral mean and variance normalization for noisy speech recognition

EURASIP Journal on Audio, Speech, and Music Processing

Table 3 Comparison of average recognition rates and percentage of improvement in comparison to MFCC for various feature types on three test sets of Aurora 2 task

Feature type	Recognition rate (%)			Percentage of improvement (%)			Overall average	Overall average improvement (%)
Feature type	Set A	Set B	Set C	Set A	Set B	Set C	Overall average	Overall average improvement (%)
MFCC	61.13	55.57	66.68	–	–	–	61.13	–
AMFCC	63.41	57.67	69.72	5.87	4.73	9.12	63.60	6.57
RAS	66.77	60.94	71.81	14.51	12.09	15.40	66.51	14.00
DAS	70.90	65.57	77.17	25.14	22.51	31.48	71.21	26.37
SPFH	73.61	68.98	80.89	32.11	30.18	42.65	74.49	34.98
MFCC-SS	69.22	63.46	73.60	20.81	17.76	20.77	68.76	19.78
MVA	76.05	76.35	73.10	38.38	46.77	19.27	75.17	34.81
ANS	77.10	74.32	83.61	41.09	42.20	50.81	78.34	44.70
Kernel	78.90	75.88	84.53	45.72	45.71	53.57	79.77	48.33
ANSS	80.47	79.04	85.53	49.76	52.82	56.57	81.68	53.05
ANS + OEP	78.78	75.98	84.14	45.41	45.94	52.40	79.63	47.92
Kernel + OEP	80.05	77.86	85.40	48.68	50.17	56.18	81.10	51.68
ANSS + OEP	82.37	81.10	86.21	54.64	57.46	58.61	83.23	56.91
ANSSOEMV	84.81	86.63	87.97	60.92	69.91	63.90	86.47	64.91
ETSI-XAFE	86.56	85.19	83.49	65.42	66.67	50.45	85.08	60.85