Synthesising Contextually Appropriate Intonation in Limited Domains

Abstract

We describe a method of synthesising contextually appropriate intonation with limited domain unit selection voices. The method enables the natural language generation component of a dialogue system to specify its intonation choices via APML, an XML-based markup language. In a pilot study, we built an APML-aware limited domain voice for use in flight information dialogues, and carried out a perception experiment comparing the APML voice to a default version built using the same recordings without the additional structure. The intonation produced by the APML voice was judged significantly more contextually appropriate than that of the default voice. These results justified building a second voice with a much larger vocabulary, using an automated script generation algorithm.