Defining the controlling parameter in constrained discriminative linear transform for supervised speaker adaptation