Real Time SpeakerRecognition on TMS320C6713

Abstract

Speaker recognition is defined as the process of identifying a person on the basis of the information contained in speech.In this world where breach of security is a major threat ,speaker recognition is one of the major biometric recognition techniques. A large number of organizations like banks,defence laboratories ,industries , forensic surveillance are using this technology for security purposes.Speaker recognition is mainly divided into two categories : Speaker identification andSpeaker verification. In speaker identification we find out which speaker has uttered the given speech ,whereas in speaker verification we determine if the speaker who is claiming a particular identity is telling the truth or not.In our first phase we did speaker recognition on MATLAB.The process we followed comprised of three parts.First we did preprocessing where we truncated the signal and performed thresholding on it.Then we extracted the features of speech signals using Mel frequency Cepstrum coefficients. These extracted features were then matched with a set of speakers using a Vector Quantization approach.In our second phase we tried to implement speaker recognition in real time.As speaker recognition is a signal processing task ,we decided to implement it real time on a DSP (digital signl processor) as it performs very fast multiply and accumulate operations(MAC) and speaker recognition had stages where signals were primarily added and multiplied .Hence DSP was choosen as our platform.The second phase comprises our familiarisation with the TMS320C6713 DSP,the first few audio applications we performed on it,some real time filters we developed on it and finally our speech recognition problem.