MFCC VQ based Speaker Recognition and Its Accuracy Affecting Factors

Abstract

The present study was conducted to evaluate the accuracy affecting factors of a Mel-Frequency Cepstral Coefficients (MFCC) and Vector Quantization (VQ) based speaker recognition system. This investigation analyses the factors that affecting recognition accuracy using speech signal from day to day life in surrounding environments. It was studied the mismatch affects of text-dependency, voice sample length, speaking language, speaking style, mimicry, the quality of microphone, utterance sample quality and surrounding noise. The corpuses of 10 people of 20 utterance subjects were collected which were indicate that any mismatch degrades recognition accuracy. It was found that most dominating factors that degrades the accuracy of speaker recognition systems were surrounding noise, quality of microphone by which voice sample were collected, disguise, and degrading of the sample rate and quality. Speech-related factors and sample length were less critical. General Terms: Speaker recognition, Speaker recognition based ATM machine, Phone banking, Database services and Man machine interface.