Speaker identification approach based on time domain extracted features

This paper presents a speaker identification approach based on features extracted by time domain speech analysis. Most features (28) issue from the TESPAR (Time Encoded Signal Processing and Recognition) coding method. The other four features are provided by the time domain analysis of the waveform. The features further employed are: the relative mean square energy, the number of maxima in the energy envelope, the pitch frequency average and the relative number of zero crossings for every utterance. This approach implies low computational requirements for features extraction and provides good recognition rates. For the experiments some classifiers (kNN, Bayes Net, Naïve Bayes, RBF and SVM) provided by the WEKA (Waikato Environment for Knowledge Analysis) environment are employed.