an input arrangement which inputs patterned features;

a base model arrangement which provides at least one base model;

an environment detector which ascertains an environment from which the at least one base model originated; and

a transform arrangement which produces a stacked target model based on a feature vector corresponding to the environment A from which the at least one base model originated using:

((ƒ_A:R^D→R^M,X′=ƒ_A(x)), wherein

A is the channel environment for training data set from which the base model originated;

X is a training data set of feature vectors of a target class;

X′ is a transformed stacked target model training set;

R^D is an input pattern feature space from the input arrangement;

R^M is a feature space calculated by base scores on the A-channel base models;

ƒ_A is the stacked target model based on the feature vector corresponding to the environment A; and

there is independence between the at least one base model and the stacked target model allowing for a single enrollment of a target class;

a second transform arrangement which produces a channel compensation stacked target model based on a feature vector corresponding to environment B using:

((ƒ_b:R^D→R^M,X′=ƒ_B(x)), wherein

B is the new channel environment;

R^D is the input pattern feature space from the input arrangement;

R^M is a feature space calculated by base scores on the set of B-channel base models; and

ƒ_B is the channel compensation stacked target model based on the feature vector corresponding to the new environment B;

a verification arrangement which compares the second transform with the first transform to arrive at a determination of mismatch in feature relationships present in environment A, wherein focus is preferably shifted from ƒ_B to or from ƒ_A depending on a degree of mismatch between the at least model from environment A and the model from environment B; and

an arrangement which produces a pattern recognition decision based on the model verification result.