Description of the systems presented:
=====================================

Common parts:
=============
Feature extraction:
	13 MFCC + 13 delta + 13 delta delta + Feature Warping

Speech Activity Detection:
	Bi-Gaussian model on frames energies to compute a threshold
	
World Model:
	Gender dependent 1024 components Gaussian Mixture Model
	
Client Models:
	Obtained from the world model using MAP adaptation
	
Baseline system UOB
===================
	z-norm depending on recording condition, i.e. normalization
	coefficients for a speaker are calculated and applied for 
	both phone and interview condition