Principal Component Analysis Based Speaker Verification
Keywords:
– Speaker verification, PCA, Feature Extraction, MFCC, EM, GMM, Feature Modeling, Vector Quantization, LBG algorithm, FCM.Abstract
Speaker verification system identifies the concern person who is speaking, through the special
characteristics of voice. Speaker verification is one to one process which is used for various safety measures purposes.
Speech/voice has some specific features (e.g., speaking style, voice pitch) which differ person to person. Throughout
verification process large speech data of concern person is not actually required. The acoustic characteristics
information is hidden in small data portion. Feature Extraction method can be applied to filter out the specific
characteristics of large data and can be store in a database after modeling. An EM Algorithm (expectationmaximization) will be used to train the data for the different uttering sounds of voice, hence that database can become
more efficient. Within the EM algorithm takes multiple iteration to calculate log likelihood value. Especially first value
of Mean is set to some random value. Setting Mean value using Fuzzy C-Mean Clustering reduces number of iteration
and increases the accuracy of result of speaker verification.After verification of input speech is to be performed with
the database, after that again Feature Extraction is done using MFCC (Mel-Frequency Cepstral Coefficients) and
modeling will be performed using GMM (Gaussian Mixture Models) on input query signal and matching will be
performed
