Modelling pause duration as a function of contextual length

Effects of contextual length are known to affect pause durations in neutral speech. The present study investigates these effects on an expressive corpus of read tales in French. Computational models of intra-sentence, and inter-sentence pause durations, as functions of contextual lengths are proposed. These models are aimed at improving Text-To-Speech synthesis systems, and provide clues for synthesizing prosodic instructions above the level of the sentence. They are also aimed to help in the prosodic analysis of pause durations, which may be biased by contextual length effects. We find the phoneme to be the best unit for measuring contextual length. Inter-sentence pause durations were more influenced by the length of the preceding sentences. Intra-sentence pause durations were more influenced by the length of the following pseudoclauses.