Voice conversion from non-parallel corpora using variational auto-encoder