Abstract
This paper proposes a method for segmentation of continuous speech into phrase-like regions for the use in broad phonetic engine. A broad phonetic engine is a system which converts input speech signal into a sequence of broad phoneme classes, namely, vowel, nasal, approximant, stop and fricative. A frame level speech/nonspeech classifier using artificial neural network is employed for detection of pause/break regions in continuous speech. Automatic marking of breaks in continuous speech is achieved using this. Performance of broad phonetic engine in Malayalam is evaluated with and without segmentation of input speech, and an improvement is obtained by segmentation. A method for searching audio database is proposed using the output of this broad phonetic engine.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call