Proper Name Pronunciations for Speech Technology Applications

摘要

This paper describes a 15-year research effort to improve the automatic pronunciation of proper names and details the issues involved in applying those pronunciations to speech synthesis and speech recognition. Our approach consists primarily of a large hand-tuned rule component, supplemented by a comparatively small pronunciation dictionary, both guided by extensive survey and polling data. Compared to other state-of-the-art programs, we use language-class identification to smaller degree. We utilize alternate pronunciations, obtained from the polling data, for both synthesis and recognition purposes. While our approach yields comparatively high accuracies, a comprehensive database of names and their pronunciations verified and authenticated through customer interactions (such as auto-attendants and automated directory assistance) will likely be the best future resource defining the ultimate in accuracy.