PLAUSIBILITY FUNCTIONS IN CONTINUOUS SPEECH RECOGNITION - THE VINICS SYSTEM

We propose a new approach to phoneme-based continuous speech recogniti on when a time function of plausibility of observing each phoneme is g iven. We introduce a criterion for best sentence, related to the sum o f plausibilities of individual symbols composing the sentence. Based o n the idea of making use of a high plausibility region to reduce the c omputation load while keeping optimality, OUT method finds the most pl ausible sentences relating to the input speech, given the plausibility mu(a,n) of observing each phoneme a at each time, slot n. Two optimiz ation procedures are defined to deal with the following embedded searc h processes: (1) find the best path connecting peaks of the plausibili ty functions of two successive symbols, and (2) find the best time tra nsition slot index for two given peaks. Dynamic programming is used in these two procedures. Since the best path finding algorithm does not search slot by slot, the recognition is highly efficient. Experimental results with the VINICS system show that the method gives a better re cognition precision while requiring about 1/20 computing time, compare d to traditional DP based methods. The experimental system obtained a 95% sentence recognition rate on a speaker-dependent test.