NOISE ROBUSTNESS OF FIRST FORMANT BANDWIDTH (F1BW) FEATURES IN MALAY VOWEL RECOGNITION

Abstract

Applications that use vowel phonemes require a high degree of vowel recognition capability. The performance of speech recognition application under adverse noisy conditions often becomes the topic of interest among speech recognition researchers regardless of the languages in use. In Malaysia, there are an increasing number of speech recognition researchers focusing on developing independent speaker speech recognition systems that use the Malay language which is noise robust and accurate. This paper present a study of noise robust capability of an improved vowel feature extraction method called First Formant Bandwidth (F1BW). The features are extracted from both original data and noise-added data and classified using three classifiers; (i) Multinomial Logistic Regression (MLR), (ii) K-Nearest Neighbors (K-NN) and Linear Discriminant Analysis (LDA). The results show that the proposed F1BW is robust towards noise and LDA performs the best in overall vowel classification compared to MLR and K-NN in terms of robustness capability, especially with signal-to-noise (SNR) above 20dB.