On the use of supra model information from multiple classifiers for robust speaker identification.