Mechanisms of Situation Adaptive Speech Production in Dialogue

Seiji NIIMI*, Shigeru KIRITANI*, Satoshi IMAIZUMI*, and Hideaki SEKI**

* Research Institute of Logopedics and Phoniatrics, Faculty of Medicine,
University of Tokyo,
** Chiba Institute of Technology

Situation dependency in dialogue speech production was investigated by analyzing read tokens (RD) and dialogue between teachers and hearing-impaired (HI) or normal-hearing (NH) children. For the tokens consisting of devoicable segments, the devoicing rate and the lengths of various segments such as moras, words, and phonological phrases were measured. Perceptual rating was also carried out to clarify the emotional profile of the tokens. Results revealed the following;

  • Mora length increased in the order of the RD, NH and HI tokens;
  • The devoicing rate decreased in the reversed order;
  • Mora length accounted for 53% of the devoicing rate;
  • Length of moras, words and phonological phrases highly correlated with emotional aspects of the tokens such as the contrast of "pleasant" vs. "discomfort";
  • Situation dependent adjustments of the temporal structure of dialogue affect not only the segmental features such as devoicing but also the prosodic characteristics which determine the emotional aspects of speech.

    Keywords: Dialogue, Temporal structure, Situation, Devoicing, Emotion