![]() |
ESCA Workshop on ProsodyLund, Sweden |
![]() |
Ten time-integrated prosodic parameters were used to assign read-out and spontaneous speech fragments to the speakers that had produced them. Fifty speakers of standard Dutch participated in the experiment. They were stratified by gender and age. It was found, that the parameters are independent of each other in terms of speaker specificity. In combination, they could correctly assign 73 % of 500 fragments to the 50 speakers. As expected, differences in mean F0 were important for speaker identification, but an analysis without mean F0 still resulted in 56 % correct classification. Identification improves when spontaneous and read-out speech are analysed separately. Thus, speech type is an important factor to control for in speaker identification.
Bibliographic reference. Kraayeveld, J. / Rietveld, A. C. M. / Heuven, Vincent J. van (1993): "Speaker specificity in prosodic parameters", In Prosody-1993, 264-267.