Fast browsing of speech material for digital library and distance learning

In digital library and distance learning applications, one usually needs to search through lots of speech material. While content-based retrieval techniques can help to identify possible matching items, the person would usually need to browse through the items quickly before making decisions on whether the items are useful or not. As a result, fast speech browsing techniques are highly desirable. In this paper we discuss problems of fast playback of speech materials and overview some existing time scale modification (TSM) techniques. We propose some novel modifications of TSM to make it much more effective in fast browsing of speech materials, especially those with irregular speech tempo. The proposed algorithm includes silent period removal, gain equalization and locally adaptive TSM. Simulation results show that the proposed algorithm can increase the intelligibility of the fast playback speech materials significantly.