Determining Prepausal Lengthening for Finnish Rule-Based Speech Synthesis

We are developing a Finnish rule-based TTS system. Our primary concern is to enhance naturalness in the synthesis by observing tendencies in natural language and implementing the findings into the synthesis. We have concentrated on modeling duration, which is essential to the Finnish language due to contrasting phonemic length and the fact that the durations of individual phones are highly sensitive to their position within a word. One of the many durational considerations is prepausal lengthening. Most languages exhibit lengthening of speech sounds at the ends of phrases and sentences, but there is no applicable data available on the phenomenon with regard to Finnish. Data mining a speech corpus, we have set out to investigate what is the best justified way to implement prepausal lengthening to a Finnish rule-based TTS system. The results show that there is considerable lengthening of the entire prepausal word. Moreover, some of the effects appear to extend to the penultimate word. The data acquired in the study provides us adequate information for modeling prepausal lengthening in TTS.