Abstract
Speaker recognition plays an important role in a biometric based identification of the person using the information available in their speech signals. In any speaker recognition system, feature extraction using signal processing approaches is an important stage. In this paper, an efficient speaker recognition system is presented by extracting the energy features of the speech signals using Discrete Wavelet Transform (DWT). Then, the extracted DWT energy features are modeled using Gaussian mixture model (GMM) classifier for the recognition of the speaker. Results prove the efficiency of the speaker recognition system with an accuracy of 96.31% at 4th level DWT features with 16 Gaussian densities.
Highlights
Speaker recognition is widely used in telephone based applications
A method of recently developed Deep Neural Network (DNN) model called as time delay DNN that is used for the large-vocabulary continuous speech recognition tasks is discussed in [3]
It investigates a lightweight change in which a supervised Gaussian mixture model (GMM) is obtained using time delay DNN posteriors
Summary
Speaker recognition is widely used in telephone based applications. There are many methods developed in the last two decades. A speaker recognition system that uses the modified Mel-Frequency Cepstral Coefficients (MFCC) method is discussed in [1]. It includes Blackman windowing which is based on different classifiers like back-propagation neural network, Euclidean distance, and self-organizing map. A method of recently developed Deep Neural Network (DNN) model called as time delay DNN that is used for the large-vocabulary continuous speech recognition tasks is discussed in [3] It investigates a lightweight change in which a supervised GMM is obtained using time delay DNN posteriors. An enhanced vector quantization algorithm is discussed in [5] for speaker recognition where the classification is attained by the results of two set of codebooks They are constructed by the use of a portion of words and the whole words respectively. It is compared to two well-known reduction scheme of dimensionality called as linear discriminant and principal component analysis
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
