Effects of Time Windowing for Extraction of Expression from Japanese Speech: Higuchi's Fractal Dimension

Measures of irregularity and complexity in timeseries data can be quantified by using fractal dimension (FD) which may convey useful information and showing a strong correlation with the physiological data, e.g., speech. The timedependent FD (TDFD) has been proven to be effective method in various physiological studies. However, the TDFD result depends on the time windowing, i.e., length and type which could provide different features. Therefore, in this paper, the effects of time windowing of TDFD by spanning 28 to 212 -point length and four window functions were investigated. The FD was computed by using Higuchi's method. In experiment, speech data recorded from 15 Japanese utterances, 4 different ways of expression ( accosting, wholehearted, normal, and uninterested). The obtained results were useful in selecting time windowing based on TDFD method in which could extract different intonations effectively.