Comparison of speaker recognition approaches for real applications