Influence of the number of principal components used to the automatic speaker recognition accuracy

Abstract

This paper discusses possibilities to reduce dimensionality of the standard MFCC feature vectors by applying the technique of Principal Component Analysis (PCA). The reported experimental results suggest that PCA is an appropriate technique to reduce dimensionality without reducing the accuracy of recognition. The applied automatic speaker recognizer shows that already for a 14-dimensional PCA feature space, the recognition accuracy reaches the target value as in the 39-dimensional MFCC feature space. This gives motivation for further research towards more efficient speaker recognizers.