Abstract
In this study, various sampling procedures were tested for F0 distributional parameters in order to see how the performance of the forensic voice comparison system is influenced. This was done by changing the width of analysis window and the degree of its shifting, resulting in different sets of feature vectors. The results show that the discriminability is fairly comparable across the different sampling procedures, whereas there is a large difference in calibration loss between the procedures if the analysis windows are not overlapped. It is reported that the use of overlapped analysis windows contributes to the stability in calibration loss. Index Terms: forensic voice comparison, long-term F0 distribution, sampling procedures
Original language | English |
---|---|
Pages | 301-304 pp. |
Publication status | Published - 2016 |
Event | Sixteenth Australasian International Conference on Speech Science and Technology - Parramatta, Australia Duration: 1 Jan 2016 → … |
Conference
Conference | Sixteenth Australasian International Conference on Speech Science and Technology |
---|---|
Period | 1/01/16 → … |