GMM-based matching ability measurement of a speech recognizer and a feature set

Abstract

In this work, we propose a Gaussian mixture model-based recognizer selection method to overcome the acoustic mismatch between training and testing environments of a speech recognition system. The method evaluates the preference of a system over other for a specific feature set. By applying it to compare the two speech recognition systems constructed with wireline speech and wireless speech, respectively, it is shown that the matched condition of wireless training and testing can give better recognition accuracies than the mismatched condition. © 2012 Springer-Verlag GmbH.