Tone is a critical characteristic in any tonal language recognition system. The Galo language, spoken in central Arunachal Pradesh, is a tonal language with limited linguistic resources. This study presents a statistical analysis of features extracted from the pitch (F0) contour of speech signals of Galo tonal words. Multivariate analysis of variance (MANOVA), t-test and mutual information were employed to identify the optimal feature sets for Galo tone recognition. Furthermore, a comparative study of feature selection algorithms was conducted using a Support Vector Machine (SVM) based tone recognition system. The SVM classifier was trained and tested using both the complete set of extracted features and the subset selected by the feature selection algorithm. Results indicated that the feature selection algorithm enhanced the performance of the SVM-based tone recognizer by 4.32%, demonstrating a significant performance improvement.

Tone is a critical characteristic in any tonal language recognition system.