Live Speaker Identification using MFCC and Delta MFCC

Speaker Identification is a subfield of digital signal processing, which verifies the identity of a person using features of their voice samples. Now days it is considered as a popular biometric verification technique in various fields. It finds application in forensic speaker recognition, authentication, surveillance and other related areas. In this paper we presented a system developed for text independent, live identification of speaker. Python 3 programming language is used to develop the system with simple tkinter GUI. Concept of Speaker Identification is inspired from the behavior of human ears. MFCC (Mel-Frequency Cepstral Coefficient) features are considered to be best for mimicking the voice signal processing in human ears. MFCC features represent only the power spectral envelope of single frames, but information of dynamic changes in the frames would be quit useful to include in the feature vector performing delta operation over features gives this information We used MFCC + Delta MFCC features of voice signal. Using the one more delta over the delta features may increase the accuracy but it increases redundancy also so we skipped that. [7]. The whole process of this technique involves basically two modules Feature extraction and feature matching. Feature extraction is a method of extracting a small amount of data/voice signal that can be used to represent each speaker. Feature matching is the process of comparing the voice inputs from a set of known speakers. We used GMM (Gaussian Mixture Model) for training and Loglikelihood function for feature matching.