USING CONTEXT DEPENDENT INFORMATION FOR DISCRIMINATIVE SPOKEN TERM DETECTION