In this paper we describe a system that uses keywords produced by a speech synthesizer embedded in carrier sentences of natural speech. This allows to generate unrestricted application-specific information, based on a set of fixed announcements. Since most of the naturalness of the utterances lies in the prosody of the carrier sentence, such a system is highly preferable to pure synthetic announcements.

Three methods to insert synthetic keywords into carrier sentences were tested and compared to pure synthetic announcements. In one method the keyword was presented in isolation with pauses to the carrier sentences. For the other methods, we modelled different kinds of coarticulation with the carrier sentence. It was found, that for most cases it is sufficient or even better to present the keywords in isolation. This procedure facilitates the technique of keyword insertion because just a correct pitch contour for the keyword must be created.