Fast Query-by-Example Spoken Term Detection Integrating the Boundary Information

This paper presents a method of fast query-by-example spoken term detection (QbE STD) integrating the phoneme boundary Information. According to this method, the phoneme posterior probabilities of query examples and test materials should be extracted firstly. and then phoneme posterior probabilities are segmented into segment sequences using hierarchical agglomerative clustering(HAC) algorithm(phoneme boundary detection), new queries and new indexes can be composed of the expectation vectors of the segment sequences. The dynamic time warping(DTW) procedure is formulated to implement QbE STD. Finally, the detection results can be modified by pseudo relevance feedback (PRF). The experimental results indicate that although the method presented by the paper has a slight reduction of the detection performance as compared with DTW, there is a great advantage over the latter in the detection speed, and compared with the method presented by other paper,the method presented by the paper also has far more superiorities in the detection speed.