ASSESSING THE IMPORTANCE OF THE SEGMENTATION PROBABILITY IN SEGMENT-BASED SPEECH RECOGNITION

The segment-based speech recognition algorithms that have been develop ed over the years can be divided into two broad classes. On the one ha nd those using the conditional segment modeling formalism(CSM), which requires the computation of the likelihood of the sequence of acoustic vectors, conditioned on the sub-word unit sequence and corresponding segmentation. On the other hand those using the posterior segment mode ling formalism (PSM), which requires the computation of the joint post erior probability of the unit sequence and segmentation, conditioned o n the sequence of acoustic vectors. The latter probability can be writ ten as the product of a segmentation probability and a unit classifica tion probability. In this paper, we focus on the role of the segmentat ion probability. After having shown that the segmentation probability is not required in the CSM formalism, we motivate its importance in th e PSM formalism. Next, we describe its modeling and training. Experime nts with two PSM-based recognizers on several speech recognition tasks demonstrate that the segmentation probability is essential in order t o obtain a high recognition accuracy. Moreover, the importance of the segmentation probability is shown to be strongly correlated with the m agnitudes of the unit probability estimates on segments that do not co rrespond with a unit. (C) 1998 Elsevier Science B.V. All rights reserv ed.