APPLYING DATA MINING CLASSIFICATION TECHNIQUES TO SPEAKER IDENTIFICATION
Kinga Saapa, Agata Trawi ́ nska, Irena Roterman‐Konieczna · 2013
Voice is one of biometric measure which can characterize an individual as unique in the whole world. Unfortunately, this assumption has not been proven so far and treating speech signal as DNA or fingerprints is not relevant. The researchers from many forensic disciplines try to find the best both signal acoustics feature(-s) and model(-s) to distinguish people via their voices. The goal of this investigation is to present effectiveness of Data Mining techniques to classification task. Four algor ithms were applied such as C&RT and CHAID classification trees and MLP and RBF neural networks models. The results show their high force to distinguish speaker. It is likely that their strength lies in abili ty to learn complex, nonlinear relations hidden in input data without any assumptions of data and model.