Word boundary detection in continuous speech based on suprasegmental features for hindi language

Abstract

Word boundary detection in continuous speech is very common and important problem in speech synthesis and recognition. Several researches are open on this field. Since there is no sign of start of the word, end of the word and number of words in the spoken utterance of any natural language, one must study the intonation pattern of a particular language. In this paper an algorithm is proposed to detect word boundaries in continuous speech of Hindi language. A careful study of the intonation pattern of Hindi language has been done. Based on this study it is observed that, there are several suprasegmental parameters of speech signal such as pitch, F0 fundamental frequency, duration, intensity, and pause, which can play important role in finding some clues to detect the start and the end of the word from the spoken utterance of Hindi Language. The proposed algorithm is based mainly on two prosodic parameters, pitch and intensity.