Maximum mutual information training of distance measures for template based speech recognition
Mike Matton, Mathias De Wachter, Dirk Van Compernolle, Ronald Cools · 2005
Previous work described an architecture for template based speech recognition in which class dependent distance measures are an important part. In this paper, a discriminative way of training these distance measures, inspired on Maximum Mutual Information (MMI), is presented. Some initial experiments with MMI training on the word-level are described. 1.