Classification of Turkish Vowels Based on Formant Frequencies

Abstract

In speech signal, formant frequency is defined as the acoustic resonance of human vocal tract. Because it is changeable across genders, age, and languages, it has been studied for various purposes by many researchers. Analyzing formant frequency of speech signal in vowels is more feasible than analyzing it in consonants. In this study, formant frequency values (F1, F2 and F3) of eight Turkish vowels (/a/, /e/, /i/, /i/, /o/, /o/, /u/, /u/) within CVC syllables have been extracted to identify most successful classification method. Totally 48 meaningful syllables in CVC context have been used for analysis and this speech corpus contains all 8 vowels existing in Turkish Language. The syllables were uttered by 10 male speakers. The results inferred that SVM (Support Vector Machines) operates best in distinguishing Turkish vowels by formant frequencies with %90 of classification success. In the future works, these findings may support studies on speech segmentation, vowel detection, speech signal enhancement and the deals that may require information about the best classification algorithm working on Turkish vowels.