Issues in generating pronunciation dictionaries for voice interfaces to spatial databases

ln speech recognition research, increasing emphasis has been placed on generating pronunciation dictionaries for spontaneous human-computer interactions. We present a review of our strategies for developing lexicons for three distinct voice interfaces: (1) recognition of proper nouns in data entry applications, (2) recognition and synthesis of spoken language for the navigation of a spatial information database using a speech-aware graphical user interface, and (3) interpretation of prosodic information to improve machine understanding performance. We discuss a broad range of issues in the development of such systems including surname pronunciations and automatic mapping of dictionary-style pronunciations into systems used to improve the performance of machine-based speech recognition and generation.