Text Independent Speaker Identification with Principal Component Analysis

Principal Component analysis (PCA) is useful in identifying patterns in data, and expressing data in a manner which highlights their similarities and differences. This concept was extracted to reduce high dimensional Mel‟s Frequency Cepstral Coefficients (MFCC) into low dimensional feature vectors. Since MFCC‟s are high in dimensions and truncation of these dependent coefficients may lead to error in identification of speaker‟s speech recognition. In this paper text independent speaker identification model is developed by combining MFCC‟s with PCA to obtain compressed feature vectors without losing much information. Generalized Gaussian Mixture Model (GGMM) was used as modeling techniques by assuming the new feature vectors follows (GGMM) [Reynolds, (1995)] [7]. The experiment was done with 40 speakers with 10 utterances of each speaker locally recorded database.