On Segmentation and Extraction of Features from Continuous Bangla Speech including Windowing

Abstract

This paper presents a speech recognition front-end system for segmenting continuously spoken Bangla speech and extracting features from the segmented words/sub-words including windows. The system commenced with recording the original speech sentences and segmenting the continuous speech into uniquely identifiable and meaningful units. After segmentation, a parametric representation has been employed for the collection of meaningful features with MFCC. Various windowing methods, such as Hamming, Hanning, Rectangular, Bohman, Trangle, Welch, Kaiser and Blackman windows were included in this feature extraction process. The test database contained 758 words/sub-words segmented from 120 sentences. Each sentence was recorded from six different speakers. Experimental results demonstrate that the developed system achieved the segmentation accuracy rate at about 95%.