Enhancement of the Input Interface of Spoken Dialogue Systems By Means of Contextual Models and Grammatical Rules

This papers proposes a new technique to enhance the performance of the input interface of spoken dialogue systems, which presents a novel contribution: the use of contextual models and grammatical rules to correct errors of automatic speech recognition (ASR). The experiments we carried out employing a spoken dialogue system shows that the technique allows enhancing word accuracy, sentence understanding and task completion by 8,5%, 16,54% y 44,17% absolute, respectively.