Feature extraction method of fractional cosine and sine transform for speaker recognition

In this paper, the novel features derived from fractional cosine and sine transform of speaker recognition task are presented, which are abbreviated to be DMEGB and Delta DMEGB. A lot of the experiments based on the multiple parameters are employed to investigate their recognition performance. The single application of above mentioned features doesn´t show the excellent results, but the reasonable combination of them which is able to describe the speaker individual characteristics more exactly will lead to the favorable effects which are even better than the conventional MFCC feature.