Using neural networks for automatic speaker recognition: a practical approach
R.G.C.P. Pinto, Herbert Laroca Pinto, L.P. Caloba · 2002
This paper describes an automatic speaker recognition method for telephone speech, using acoustic processing of a single trisyllabic Portuguese word. Cepstrum coefficients are extracted from the linear predictive coefficients and no further processing is needed. In this method, the traditional classification method (non-Euclidean distance measurement) was efficiently substituted by a multi-layer neural network with encouraging results.