Learning Representations of Emotional Speech with Deep Convolutional Generative Adversarial Networks