Proposed combination of PCA and MFCC feature extraction in speech recognition system

In speech recognition system, the Mel Frequency Cepstrum Coefficients (i.e. MFCC) feature extraction is an important process. It has also been wildly used in many applications. In this paper, we present the conventional MFCC feature extraction method and propose two novel versions of MFCC method that will combine the PCA technique and conventional MFCC feature extraction method. Finally, these three different MFCC methods will be tested in terms of recognition accuracy and the execution time of the HMM training process. From these two measures (i.e. recognition accuracy and time complexity of HMM training process), the developers can choose the appropriate MFCC method for the speech recognition application.