Enhancing automatically discovered multi-level acoustic patterns considering context consistency with applications in spoken term detection.