In this paper we present improvements of a frame-synchronous noise compensation algorithm that uses Stochastic Matching approach to cope with time-varying unknown noise. We propose to estimate a hierarchical mapping function in parallel with Viterbi alignment. The structure of the transformation tree is build from the states of acoustical models. The objective of this hierarchical transformation is to better compensate non-linear distortions of the feature space. The technique is entirely general since no assumption is made on the nature, level and variation of noise. Our algorithm is evaluated on the VODIS database recorded in a moving car. For various tasks, proposed technique significantly outperforms classical compensation/adaptation methods.