Analysis of speaker's gender effects in voice onset time of Turkish stop consonants

Abstract

In speech processing, identification of speaker's gender has been considered as a topic of interest by many studies so far. Although various methods and approaches have been proposed to detect gender of a speaker, Voice Onset Time (VOT) values of stop consonants of a language can also contain cues about the speaker's gender. In this study, VOT values of six stop consonants (/p/, /b/, /t/, /d/, /k/, /g/) in Turkish language have been analyzed to find gender evidences by using classification algorithms like SVM, KNN, Logistic Regression etc. 10 males and 10 females speakers have uttered 48 isolated meaningful (number of combinations of all stop consonants and all vowels) Turkish syllables in this work. As a result of this study, VOT values of unvoiced stop consonants (/p/, /t/, /k/) together as one parameter have been found to be gender-related with a success rate of 90% via Ensemble algorithm. Also, stop consonant /t/ has been marked as most gender-related cue in a syllable, on the other hand /b/ has been detected as the least gender-related stop. These findings proved that acoustic VOT parameters of Turkish stop consonants, especially unvoiced stops, can be seen as gender distinctive feature for Turkish speakers.