5th International Conference on Spoken Language Processing
For the purpose of constructing prosodic rules for dialogue speech synthesis, a comparative study was conducted on speech rates between dialogue speech and read speech. Based on the generation modeling of F0 contour, we can define 4 prosodic units, such as prosodic sentence, prosodic phrase and so on. Speech rate was analyzed with respect to these units. In a prosodic sentence, dialogue speech starts with a speech rate similar to that of read speech. The speech rate then gradually increases and, after passing through the middle of the unit, decreases towards the end. Similar tendencies were also observed in lower level units, but the degree of speech rate change in a unit was smaller. Based on the above results, a preliminary rules for speech rate control were developed for dialogue speech synthesis. A hearing test showed that the developed rules could make the synthetic speech sound more dialogue-like.
Bibliographic reference. Hirose, Keikichi / Kawanami, Hiromichi (1998): "On the relationship of speech rates with prosodic units in dialogue speech", In ICSLP-1998, paper 0730.