A Novel Approach to Design ASV Based Security System Using Neural Network

摘要

Traditional password/token based authentication schemes are insecure and presently are being replaced by biometric authentication mechanisms. Voiceprints were one of the first biometrics to be automated to use in authentication schemes. The main aim of this paper is to develop an Automatic Speaker Verification (ASV) System using Neural Networks which can be used in security purpose, criminal identification etc. Information extraction is done from the voiceprints via Modified Mel-Frequency Cepstral Coefficients (MFCC) Method. A typical ASV system has been designed for five persons and its performances under realistic physical conditions are examined thoroughly. Results are presented in this paper in the form of necessary graphs and plots.