Text this: Unsupervised feature learning for speech using correspondence and Siamese networks