Speaker recognition over the course of 10 years and across speech styles

Gráczi, Tekla Etelka and Fejes, Attila and Krepsz, Valéria and Huszár, Anna (2022) Speaker recognition over the course of 10 years and across speech styles. ALKALMAZOTT NYELVTUDOMÁNY (SI). pp. 94-109. ISSN 1587-1061 (nyomtatott), 2498-4442 (online)

Preview

Text 06_Graczi_Fejes_Krepsz_Huszar.docx.pdf Download (1MB) | Preview

Abstract

A common task in forensic speaker recognition is to determine whether two speech samples recorded over a long period of time, perhaps years apart, are likely to be the same. The present study addresses one aspect of this problem. We have analyzed the possibility of speaker recognition in recordings made ten years apart. Speech samples of various speech types were used from 6 female and 14 male speakers. Each speaker was recorded twice, 10 years apart, using a matching protocol. The speech samples were compared using a voice biometrics system (VOCALISE), a reportedly highly efficient software (for close timed recordings). The resulting score data were analyzed using linear mixed models by the means of R. The results show that the recordings made 10 years apart have influenced the identification results. However, there is a large difference in scores between speakers. Comparing different types of speech samples, we found that the speech task played a more important role than the speech style. This difference is negligible in samples recorded ten years later, while it plays an important role within the same recording sessions.

Actions (login required)

Edit Item