Unsupervised speech representation learning using WaveNet autoencoders