Fusion of phonotactic and prosodic knowledge for language identification