Automatic metric-based speech segmentation for broadcast news via principal component analysis.

摘要: In this paper, we proposed an algorithm used to improve the performance of metric-based segmentation techniques, by which points are found at maxima a distance measured between two contiguous windows shifted along stream speech features. our method, PCA processes first performed on features obtain more robust features, and then above was applied PCA-derived decide points. Experiment results show that method can efficiently detection rates up 7% while false alarm remain unchanged.