Experimental evaluation of a new speaker identification framework using PCA

In a speaker identification system, training speaker models (e.g. Gaussian mixture model, GMM) is computationally expensive, especially when the dimension of feature vectors is large. Principal component analysis (PCA) method is an optimal linear dimension reduction technique in the mean-square sense, which can reduce the computational overhead of the subsequent processing stages. In this paper, a new speaker identification framework is proposed, with PCA embedded in after feature extraction step. Experiments are conducted to investigate PCA de-correlation and dimension reduction properties. The robust ability of PCA transform is also examined. Some promising results are found.