Exploiting sparsity in Expectation-Maximization for fitting finite mixture models