Towards fully Bayesian speaker recognition: integrating out the between-speaker covariance | Publicación