国际会议最佳学生论文奖(Best student paper).pdfVIP

CRSS SYSTEMS FOR 2012 NIST SPEAKER RECOGNITION EVALUATION Taufiq Hasan, Seyed Omid Sadjadi, Gang Liu, Navid Shokouhi, Hynek Boril, John H.L. Hansen ˇ Center for Robust Speech Systems (CRSS) Erik Jonsson School of Engineering Computer Science The University of Texas at Dallas (UTD), Richardson, TX 75080-3021, USA ABSTRACT capability and calibration performance of the systems can be evalu- ated. The following aspects are considered: (a) the test utterances This paper describes the systems developed by the Center for used in these two tasks are non-overlapping; (b) the Dev-Test and Robust Speech Systems (CRSS), for the 2012 National Institute Dev-Train utterances are also included in Eval-Train so that the lat- of Standards and Technology (NIST) Speaker Recognition Evalu- ter contains the maximum amount of enrollment data per speaker; (c) ation (SRE). Given that the emphasis of SRE’12 is on noisy and training and test segments have different Linguistic Data Consortium short duration test conditions, our system development focused on: (LDC) labels (LDC-ID) ensuring a channel mismatch; (d) held out (i) novel robust acoustic features, (ii) new feature normalization speakers’ data from SRE’06 are included in tests to serve as unknown schemes, (iii) various back-end strategies utilizing multi-session non-target speakers; (e) both telephone and interview speech is used and multi-condition training, and (iv) quality measure based system for enrollment