Spectral distance measures for matching consecutive speech segments

In this study, spectral distance measures are compared for the task of matching consecutive speech segments. A new distance measure is proposed for this task, which decreases the error of the segment matching system from %16.6-22.1 to %9.8-12.5. A decision refinement algorithm based on pitch period estimation and LPC parameters is developed. The error of the system is reduced to %8.6-9.8 by using this algorithm.