Pairwise Learning Using Unsupervised Bottleneck Features For Zero-Resource Speech Challenge 2017 (System 1)

The system is for track1 alone. We trained an antoencoder using unsupervised bottleneck features with word-pair information from Switchboard. The unsupervised bottleneck features was extracted from an extractor of multi-task learning deep neural networks (MTL-DNN). The word-pair information was the ground truth from Switchboard. The final features are obtained from the third layer in our pairwise trained autoencoder.