An analysis of acoustic deviation manner in spontaneous speech

Natural speech typically contains various phenomena deviated from the formal mode such as read speech. It is well known that those paralinguistic phenomena have an important role to give the human emotions and the state of the speakers in speech communication. This study attempts to extract the deviation as an acoustic ‘‘vagueness,’’ defined by temporal and dynamical acoustic features of speech. Especially the change of the vagueness during a certain period of speech, such as a 10‐minute presentation, is focused. As the acoustic features, it used (i) modulation spectrum and (ii) syllable speed, which may have relations to the speech clarity and the tempo. For the experiments, 70 academic presentation speech data in the Corpus of Spontaneous Japanese (CSJ) are used. As the experimental results, significant properties in the patterns of the modulation spectrum and the syllable speed are obtained as a difference of the beginning and the ending periods of the presentation. This result will contribute to a humanlike speech dialog system.