Current linguistic theory concerning intonation and prosodic structure is often predominantly based around isolated examples of `interesting phenomena', which may be linguistically interesting, but only account for a small percentage of utterances that the average TTS system would be required to speak. These theoretical and recorded examples also tend to be short or over simplistic compared to the kind of utterances which are found in the various styles of naturally occurring speech that a TTS system would be require to mimic.To improve intonation in speech synthesis, it is the general structure of intonation that is the primary concern here. With this end in mind, this paper presents findings from the analysis of a corpus of read news articles, showing what changes occur within the speaker's pitch range, and tries to relate these findings to current linguistic ideas.
To download this paper, please return to Proceedings of the 1998 Postgraduate Conference