Furcax: End-to-end Monaural Speech Separation Based on Deep Gated (De)convolutional Neural Networks with Adversarial Example Training.