Özellik vektörlerinde enerji türevleri ile konuşmacı bağımsız Türkçe konuşma tanıma iyileştirmesi

Abstract

At the recent times, speech recognition applications, which are increasingly used in smart devices, are gaining importance as they perform well on speaker-independent systems. In this study, an increase is obtained in the feature vectors extracted from the speech by traditional studies. For this process, energy and delta derivatives were applied to measure hidden speaker identities in audio bands. The coefficients obtained are given as input values to gain estimation ability to intelligent devices. The comparison and evaluation of the work, that is done with the traditional studies, is presented to indicate the efficiency of the work. © 2018 IEEE.