Parametric study on speaker identification biometric system using formant analysis.
Binsu C. Kovoor, M. H. Surpiya, K. Poulose Jacob · The Journal of the Acoustical Society of America · 2009
Speech is the most suitable bio-feature for environments where hands free is a requirement in the security system. Speaker identification is done successfully utilizing formant frequencies in the human speech. In the prototype system developed, which can be operated without the knowledge of the users, representative features are extracted and compared with speaker specific features stored in the knowledge base. The algorithm mainly consist of four modules, namely, recording of samples and building a knowledge base of the vowel sounds, building autoregressive models, recognizing vowel and thereby identifying the speaker. The important parameters to be fixed for obtaining optimum output for a given input data of the proposed model are envelope threshold and order of parametric AR model. The envelope threshold is the parameter that differentiates numerous peaks in the speech signal while an appropriate value of the order of AR model helps to compute the frequency response of the input signal accurately. The influences of various parameters have been studied and are discussed in this paper. It is expected that the results of the study will help a biometric system developer to choose appropriate value for the parameters in the algorithm for different applications.