Software Open Access
Sanderson, Conrad; Curtin, Ryan
Armadillo C++ library v7.960 containing gmm_diag and gmm_full classes. The gmm_diag and gmm_full classes provide multi-threaded (parallelised) implementations of Gaussian mixture models and the associated Expectation-Maximisation training algorithm. The gmm_diag class is specifically tailored for diagonal covariance matrices (all entries outside the main diagonal in each covariance matrix are assumed to be zero), while the gmm_full class is tailored for full covariance matrices. The gmm_diag class is typically much faster to train and use than the gmm_full class, at the potential cost of some reduction in modelling accuracy.