Abstract
A laboratory experiment was done to test speech recognition accuracy when the template patterns, against which spoken utterances were matched, were generated by combining the speech of several different talkers. This allowed several different speakers to talk in rapid succession to the system without each having to load his or her patterns. In general, median word recognition accuracy decreased by a few percent with composite patterns. The amount of the decrease depended on the size and composition of the group. This was considered to be a tolerable decrease for some applications. As the experiment progressed, it became apparent that the more difficult problems were human factors issues related to error handling, vocabulary choice, user groups, etc. These are discussed.

This publication has 2 references indexed in Scilit: