Analysis of throat microphone using MFCC features for speaker recognition

Abstract

In this paper, a visual aid system has been developed for helping people with sight loss (visually impaired) to help them in distinguishing among several speakers. We have analyzed the performance of a speaker recognition system based on features extracted from the speech recorded using a throat microphone in clean and noisy environment. In general, clean speech performs better for speaker recognition system. Speaker recognition in noisy environment, using a transducer held at the throat results in a signal that is clean even in noisy. The characteristics are extracted by means of Mel-Frequency Cepstral Coefficients (MFCC). Radial Basis function neural network (RBFNN) and Auto associative neural network (AANN) are two modeling techniques used to capture the features and in order to identify the speakers from clean and noisy environment. RBFNN and AANN model is used to reduce the mean square error among the feature vectors. The proposed work also compares the performance of RBFNN with AANN. By comparing the results of the two models, AANN performs well and produces better results than RBFNN using MFCC features in terms of accuracy.