Tid: Onsdag 25 maj 2016, kl. 15:00-17:00
Plats: Övningslaboratoriet, plan 2, hus C, Södra huset, Frescati

Postseminarium följer direkt efter seminariet i institutionens pentry.​

In this talk we present a method for estimating the minimum speech sample length necessary to derive representative values of long-term measures of fundamental frequency central tendency. It consists of applying change point analysis to a time series comprised of values of a central tendency measure calculated cumulatively along an F0 contour. The analysis finds the time-point that separates the time series in two parts having statistically different underlying variances, a point we call stabilization point. Three central tendency measures are tested: arithmetic mean, median and the baseline. We present two experiments designed to test the effect of language, speaking style and speaker on stabilization times of the three measures. The overall results show that the three measures stabilize around 10 seconds, the baseline having slightly faster times. In some languages speaking style can have an effect on stabilization time variability, although variation among speakers and languages does not seem to have an effect on stabilization times. ​

Hjärtligt välkomna!

Anders Eriksson och Hatice Zora