Title :
Forensic Identification Reporting Using A GMM Based Speaker Recognition System Dedicated to Algerian Arabic Dialect Speakers
Author :
Bengherabi, Messaoud ; Tounsi, Bilal ; Bessalah, Hamid ; Harizi, Farid
Author_Institution :
Centre de Dev. des Technol. Av., Algiers
Abstract :
Starting from the fact of the lack of Arabic databases dedicated to performance evaluation of speaker recognition and forensic reporting systems. We present in this paper our experience in constructing an Algerian dialect database and the motivation beyond this work. After that, the corpus based Bayesian framework for interpretation of evidence in forensic systems in terms of likelihood ratio (LR) is applied on this database under different recording conditions: microphone, fixed and cellular. Preliminary results in terms of Receiver Operating Characteristics (ROC) and TIPPET plots show higher accuracy in matched conditions of training and testing. However, the performance degrades significantly in mismatched conditions.
Keywords :
Bayes methods; Gaussian processes; audio databases; biometrics (access control); maximum likelihood estimation; police data processing; speaker recognition; Algerian Arabic dialect speakers; Algerian dialect database; GMM based speaker recognition system; corpus based Bayesian framework; forensic identification reporting systems; likelihood ratio; receiver operating characteristics; Bayesian methods; Biometrics; Databases; Degradation; Forensics; Microphones; Natural languages; Speaker recognition; Speech; Testing; Gaussian mixture model; TIPPET plot; component; forensic reporting; mismatched recording conditions; strength of evidence;
Conference_Titel :
Information and Communication Technologies: From Theory to Applications, 2008. ICTTA 2008. 3rd International Conference on
Conference_Location :
Damascus
Print_ISBN :
978-1-4244-1751-3
Electronic_ISBN :
978-1-4244-1752-0
DOI :
10.1109/ICTTA.2008.4530024
