Detection of Large Segmentation Errors with Score Predictive Model

摘要

This paper investigates a possibility of an utilization of regressive score predictive model (SPM) in a process of detection of large segmentation errors. SPM's scores of automatically marked boundaries between all speech segments are examined and further elaborated in an effort to discover the best threshold to distinguish between small and large errors. It was shown that the suggested detection method with a proper threshold can be used to detect all large errors for a specific type of a boundary.