An iterative deep learning framework for unsupervised discovery of speech features and linguistic units with applications on spoken term detection.