DeepMind's latest paper contain the method for learning latent space using vector embedding.
With this method, the neural network can learn phoneme without supervision.
Anyone who are interesting about this paper, please click the below url.
Paper URL: https://arxiv.org/pdf/1711.00937.pdf
You can hear some samples related with speech in the below blog. The samples are playing encoded speech using Wavenet decoder or changing voice style combining speaker id.
