Abstract
All monologue speech data in the Corpus of Spontaneous Japanese come with subjective impression rating scores regarding the way the monologue was spoken (fluent/disfluent, expressive/monotonous, confident/not-confident, relaxed/nervous and so on), an impression of the speakers' voice quality, and speaking rate. The relation among various evaluation keywords was analyzed using Hayashi's quantification type III (Principal Component Analysis for categorical data). The first two dimensions extracted from the pooled evaluation data of more than 1400 samples were interpreted as corresponding to an overall impression of "Positive-Negative" (the first dimension) and "Activity" (the second dimension). Multiple regression analysis between the scores along the first dimension and various characteristics extracted from speech samples revealed a high correlation between the "Fair-Poor" score and the pause-ratio of samples.
