Prosody-Based Sentence Boundary Detection of Spontaneous Speech

Sentence boundary detection (SBD), also known as sentence breaking decides where a sentence begins and ends. This paper describes sentence boundary detection using acoustic and prosodic features for spontaneous Malay language spoken audio. We introduced the addition of volume change rate to 7 prosodic features and rate-of-speech for our preliminary experiment of detecting sentence boundary. Experiments are conducted on a forty-two minutes question-answer (Q/A) session of spontaneous speech comprising 12 adult male speakers and 4 female speakers. The speech datasets are first classified as speech/non-speech segments and only the non-speech segments are further tested as candidates of sentence boundaries. Our proposed rule-based method of boundary detection managed a promising 74.88% accuracy rate. For future work, we are considering to utilize learning algorithm to improve the accuracy rate and reduce false alert.