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 2 Comparison of overall average recognition rates of different number of MFCC features with Fisher-Markov selector on three test sets of Aurora 2 task

Number of MFCC coefficients	Recognition rate (%)			Overall average of recognition rate (%)
Number of MFCC coefficients	Set A	Set B	Set C	Overall average of recognition rate (%)
3 MFCC + log energy + their first and second order derivatives	57.38	44.95	51.29	51.21
4 MFCC + log energy + their first and second order derivatives	58.48	46.25	52.65	52.46
5 MFCC + log energy + their first and second order derivatives	60.41	48.16	54.46	54.34
6 MFCC + log energy + their first and second order derivatives	60.27	48.04	54.34	54.22
7 MFCC + log energy + their first and second order derivatives	62.49	50.13	56.45	56.36
8 MFCC + log energy + their first and second order derivatives	64.87	52.98	58.31	58.72
9 MFCC + log energy + their first and second order derivatives	64.23	52.01	57.48	57.91
10 MFCC + log energy + their first and second order derivatives	65.24	53.18	59.12	59.18
11 MFCC + log energy + their first and second order derivatives	64.56	52.63	57.98	58.39
12 MFCC + log energy + their first and second order derivatives	65.96	54.78	60.32	60.35