A novel segmentation method of Sound-Packets for Bangla speech signal

This paper describes several Sound-Packet segmentation techniques, which will facilitate Automatic Speech Recognition (ASR) for Bangla speech signal. The approximate duration of a sound-packet has been determined and an envelope-detection method has been presented to determine the end-points of sound-packets. The 1^st difference method, based on moving average of 1^st difference of the signal, is then introduced. Finally, the Mel-Frequency Cepstral Coefficients (MFCC) are determined and have been applied along with other techniques for successful segmentation of Bangla sound-packets.