Simulating spontaneous speech: application to forensic voice comparison

Yuko Kinoshita, Michael Norris

Research output: A Conference proceeding or a Chapter in BookConference contribution

Abstract

We investigated the feasibility of improving the reliability of LR calculations by creating a simulated background population. Using spontaneous speech and carefully read-out speech from 18 male speakers of Australian English in the ANDOSL database, we modelled the characteristics of the two speech styles and applied the model to Bernard’s formant data to simulate spontaneous speech data from it. We calculated two sets of LRs using two different background populations: Bernard’s original data, and the simulated spontaneous speech data. The simulated spontaneous speech data did not perform better in the classification of speaker pairs, but did improve the calibration of the LRs significantly.
Original languageEnglish
Title of host publicationProceedings of the 13th Australasian International Conference on Speech Science and Technology
Place of PublicationMelbourne, Australia
PublisherAustralian Speech Science and Technology Association (ASSTA)
Pages26-29
Number of pages4
ISBN (Print)9780958194631
Publication statusPublished - 2010
EventSST 2010: Thirteenth Australasian International Conference on Speech Science and Technology 2010 - Melbourne, Australia
Duration: 14 Dec 201016 Dec 2010

Conference

ConferenceSST 2010: Thirteenth Australasian International Conference on Speech Science and Technology 2010
CountryAustralia
CityMelbourne
Period14/12/1016/12/10

Cite this

Kinoshita, Y., & Norris, M. (2010). Simulating spontaneous speech: application to forensic voice comparison. In Proceedings of the 13th Australasian International Conference on Speech Science and Technology (pp. 26-29). Australian Speech Science and Technology Association (ASSTA). http://assta.org/sst/SST-10/SST2010/PDF/AUTHOR/ST100035.PDF