HIDDEN MARKOV MODEL BASED BROADCAST NEWS SEGMENTATION AND CLASSIFICATION

A new HMM-based segmentation and classification algorithm is proposed for the segmentation and classification of broadcast news since HMM can simulate stochastic time series data quite well. Firstly, by using an HMM, which has two hidden semantic states, the raw broadcast news is coarse-grained segmented into two parts: prelude/finale and speech. Then three HMMs are used to pre-classify speech clips as anchorpersons, commercials and weather forecasts based on maximum probability. Finally the change of semantic rate is checked to identify the detailed report.