Filler words as a speaker classification feature

要旨

Many of us can think of iconic words or expressions that our close friends or family often use. This study investigated the potential of idiosyncrasies in usage of filler words as a speaker classification parameter, using spontaneous Japanese speech. We estimated likelihood ratios (LRs) based on which filler words were used and how often, and performed speaker discrimination tests using the LRs. As a result we discovered that speakers� choices of fillers can predict speaker�s identity to some degree, and that this has potential as an additional feature for speaker classification.

主題