Hybrid Models In Speaker Identification