Continuous Bangla Speech Segmentation, Classification and Feature Extraction
Md. Mijanur Rahman, Farukuzzaman Khan, Jatiya Kabi, Kazi Nazrul · 2012
Continuous speech recognition is a multileveled pattern recognition task, which includes speech segmentation, classification, feature extraction and pattern recognition. In our work, a blind speech segmentation procedure was used to segment the continuously spoken Bangla sentences into words/sub-words like units using the endpoint detection technique. These segmented words were classified according to the number of syllables and the sizes of the segmented words. MFCC signal analysis technique was used to extract the features of speech words, which including windowing. The developed system achieved the segmentation accuracy rate at about 98 % and total 24 sub-classes of segmented words with MFCC features.