Uso de información contextual en la interfaz de entrada de un sistema de diálogo

Resumen

This paper proposes a new technique that aims to enhance the performance of the input interface of spoken dialogue systems by means of combining the advantages of the prompt-dependent and prompt-independent grammars used by these systems. The technique proposes creating a two-level speech recogniser since it contains two modules. The first one is a continuous-speech standard recogniser that receives the user voice and produces as output a graph of words. The second one is a word-graph analyser that processes the output of the first module considering the prompt generated by the dialogue system. The input interface of a spoken dialogue system has been evaluated using the technique. The experimental results show that when the in-context utterances are analysed, the speech recognition and understanding rates increase in 4.09% and 4.19%, respectively. They also show that when the out-of-context utterances are analysed, the speech recognition and understanding rates increase in 87.74% and 82.04%, respectively.