Using New Hybrid Filter for Selecting Features in Speaker Recognition Systems

this paper proposes a new filter for Selecting Spectral Features in Speaker Recognition Systems. The goal of algorithm is to increase accuracy of Speaker Detection System by using Supported Vectors Machine as a Classifier. First we use ACO1 and Genetic method for selecting optimized subset then present a new hybrid filter for selecting MFCC 2 features and modify SVM3 classifier to reach better accuracy for Detecting Speakers. Resultsshow that the Error Rate, number of selected features and time complexity of the proposed method is lower than other methods.

Keywords:

ACO method , Genetic method , , MFCC feature Spectral Features , Speaker Detection Systems , SVM