A new word separation algorithm for continuous Bangla Speech Recognition

摘要

Word separation algorithm is needed to separate words from continuous speech for improving performance of Speech Recognition. Studies show that existing techniques suffer from separating words into disjoint sub-words and then we present a new word separation algorithm to overcome this problem. Prosody has great impact on Bangla speech and the algorithm is developed by considering prosodic feature with energy. At first continuous Bangla speech signals are fed into the system and the word separation algorithm separate speech into isolate words. Performance of the proposed algorithm is compared to the existing algorithms and result shows that 98% word boundaries are correctly detected.