Integrating frame-based and segment-based dynamic time warping for unsupervised spoken term detection with spoken queries