- The Dempster-Shafer Theory for Statistical Model-Based Voice Activity Detection and Method for Improving Performance thereof

摘要

Disclosed are a statistical model-based voice detector based on a Dempster-Shafer theory (DST), capable of withdrawing the final voice existence probability, and a performance improvement method thereof. According to the present invention, the method comprises the following steps: finding each likelihood ratio from a plurality of statistical model-based voice detectors based on a complex Gaussian distribution; using each likelihood ratio to withdraw each voice existence probability; modifying each voice existence probability into basic probability assignment (BPA) considering each entropy; and combining a plurality of modified BPAs to withdraw the final voice existence probability, and to use the same in voice detection.