Improving speech endpoint detection using fuzzy logic-based methodologies

A novel approach to the development of an efficient speech endpoint detector (EPD) has been proposed to be applied to quasi-continuous and continuous speech processing. The goal is the improvement of multiple-threshold traditional EPDs to fit a larger applications range. Fuzzy logic processing has been applied to implement the decision logic needed to evaluate uncertain data related to transition points among contiguous speech units. Results of the fuzzy logic-based EPD has been compared to the multiple threshold-based EPDs. Extensions towards phonetic unit segmentation and a practical application in virtual agents animation have been discussed.