Formant Based Bangla Vowel Perceptual Space Classification Using Support Vector Machine and K-Nearest Neighbor Method
Sourin Dey, Md. Ashraful Alam · 2018
In the emerging field of speech processing and Automatic Speech Recognition (ASR), vowel perceptual space classification has a vital role for speech intelligibility. In this paper, formant based vowel perceptual space classification is implemented for Bangla vowels. A dataset of vowel signals for 50 speakers has been prepared. The first and second formants of vowels have been extracted from segmented recorded data of different speakers. These two formants have been employed to classify the Bangla vowels perceptual space. Two algorithms namely Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) are used to classify the vowels perceptual space using formants. SVM linear kernel has turned up to be efficient with 84.3% classification accuracy and SVM radial basis function (rbf) kernel has shown to be 100% accurate. KNN has exhibited maximum of 95% classification accuracy.