Multilingual Speech Interfaces for Resource-Constrained Dialogue Systems

Abstract

A key component for the successful provision of spoken dialogue systems is speech recognition. The capabilities of the applied speech recognizer influence many other design criteria for spoken dialogue systems. In this chapter, the problem of multilingual speech for speech recognizers is analyzed and a solution for recognizing many languages simultaneously is proposed. Two main design goals for the described system were to keep the recognition feasible on embedded systems and to handle the native speech of the user with highest priority. In addition, experiments are described that address the effect of non-native accent. With the help of the added multilinguality, existing in-car infotainment systems can be extended to international navigation input and music selection via voice.