| Summary: | 碩士 === 中華技術學院 === 電子工程研究所碩士班 === 96 === The Mel-scale frequency cepstral coefficients (MFCC) are the popular coefficients to be used in speaker recognition and speech recognition. The procedures to obtain the Mel-scale frequency cepstral coefficients are: framing and filtering the speech data by Mel-scale cepstrum filter bank, having the logarithmic energies of the output of the filters, obtaining the feature parameters of speeches by using Discrete Cosine Transformation (DCT) operation. In this study, the coefficients of the linear prediction error filters are obtained in the first. Then, with the obtained linear prediction coefficients, the linear prediction derived cepstral coefficients (LPCC) are obtained as the feature parameters.
Experimental results show that the performances of speaker recognition are very similar between the method using MFCC and the method using LPCC.
|
|---|
