SPEAKER-CONSISTENT PARSING FOR SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION

This paper describes a novel speaker-independent speech recognition me thod, called ''speaker-consistent parsing'', which is based on an intr a-speaker correlation called the speaker-consistency principle. We foc us on the fact that a sentence or a string of words is uttered by an i ndividual speaker even in a speaker-independent task. Thus, the propos ed method searches through speaker variations in addition to the conte nts of utterances. As a result of the recognition process, an appropri ate standard speaker is selected for speaker adaptation. This new meth od is experimentally compared with a conventional speaker-independent speech recognition method. Since the speaker-consistency principle bes t demonstrates its effect with a large number of training and test spe akers, a small-scale experiment may not fully exploit this principle. Nevertheless, even the results of our small-scale experiment show that the new method significantly outperforms the conventional method. In addition, this framework's speaker selection mechanism can drastically reduce the likelihood map computation.