Discriminative Articulatory Feature-Based Pronunciation Models With Application To Spoken Term Detection