Utd-crss Systems for Nist Language Recognition Evaluation

This study summarizes the overall solution and sub-systems developed by the Center for Robust Speech Systems (CRSS) at the University of Texas at Dallas to address the NIST LRE-2011 competition. CRSS-UTD employs five core sub-systems in the proposed language ID solution that include: (1) i-vector, (2) SVM-GSV, (3) PPRLM, (4) Articulatory Feature based, and (5) Prosody based. The first four represent the core solutions, and the fifth represents an investigative effort to incorporate micro-prosodic structure which has previously been explored for dialect ID by CRSS [1], [2]. This paper is organized as follows: first (i) data preparation is discussed for the system development; followed by (ii) System Descriptions; and finally (iii) probe results using LRE-09 data. Score combination of the resulting sub-systems was also considered for overall system development.