Skip to Main Content
A procedure utilizing linear threshold techniques for segmenting utterances of a single known phrase has been implemented. The technique utilizes several weight vectors derived in an iterative manner and temporal information between various designated speech events to perform segmentation. Accuracy results ranging from 100 to 92 percent have been achieved on nondesign data and are considered to be encouraging. Some remarks are included concerning means for strengthening these techniques and for applying them to other areas.