A dialog control strategy based on the reliability of speech recognition

This paper considers a dialog control strategy based on the reliability of speech recognition. The spoken dialog system using this strategy accepts an utterance of the user if the reliability is high, while it rejects the utterance and asks the user to speak again (the basic strategy), or confirms the content of the utterance if the reliability is low. The purpose of this paper is to estimate two quantities P_ac and N, given the performance of the speech recognizer used in a dialog system. P_ac is the probability that information included in the user´s utterance is conveyed to the system correctly, and N is the average number of turns taken between the user and the system until the subdialog on the user´s first utterance terminates. The analysis has proven that the direct confirmation can increase P_ac and the indirect confirmation can reduce N in comparison with the basic strategy. The paper also reported a numerical evaluation of the dialog control strategy conducted by using a real speech recognition system