Fast likelihood computation for continuous-mixture densities using a tree-based nearest neighbor search