We present the automatic forensic voice comparison system SPES which is developed in cooperation with the Federal Criminal Police Office of Germany (Bundeskriminalamt - BKA), the University of Applied Sciences Koblenz and the Department of Phonetics, University of Trier, Germany. The system is based on the classical GMM-UBM framework based on MAP adaption as described by Reynolds et al. This GMM-UBM framework is extended by creating specific background models for each suspect recording based on the similarity of speakers. The less general and more specific background models lead to better discrimination ability and more reliable results. We show how the usage of recording adapted background models outperforms the standard GMM-UBM approach in different recording conditions.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=15497
(2337KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
