Improved Malay Vowel Feature Extraction Method Based on First and Second Formants

There are many speech recognition applications that use vowels phonemes. Among them are speech therapy systems that improve utterances of word pronunciation especially to children. There are also systems that teach hearing impaired person to speak properly by pronouncing words with a good degree of intelligibility. All of these systems require high degree of vowel recognition capability. This paper presents a new method of Malay vowel feature extraction based on formant and spectrum envelope called First Formant Bandwidth (F1BW). It is an effort to increase Malay vowel recognition capability by using a new speech database that consist of words uttered by Malaysian speakers from the three major races, Malay, Chinese and Indians. Based on single frame analysis, F1BW performs better than MFCC by more than 9% based on four classifiers of Levenberg-Marquart trained Neural Network, K-Nearest Neighbours, Multinomial Logistic Regression and Linear Discriminant Analysis.