Speaker identification based on kernel Ho-Kashyap classifier

This paper presents application of kernel Ho-Kashyap KHK classifier to speaker identification. Since this classifier achieves very good performance for most benchmark data sets used in machine learning, it was very interesting to check its performance in speaker identification which is a slightly different problem. Comparison of achieved identification accuracy with popular modeling techniques in speaker recognition like Gaussian mixture models and vector quantization is made. All research is based on Polish speech corpus ROBOT, which is designed for testing speech algorithms. Identification results are discussed.