The main problem of state of the art systems for automatic speech recognition (ASR) is their lack of robustness, manifested as a substantial performance degradation in the presence of additive noise and/or linear filtering. The first step in all speech recognition algorithms is to represent consecutive speech segments of speech using low-dimensional feature vectors, the purpose of which is to remove any variability and redundancy which is believed to be irrelevant to the recognition. However, it is not clear that in this process of dimension reduction, and peeling off of what seems to be speech component unnecessary for recognition, one is not discarding the information which makes speech such a robust message representation. In this talk we will address the issue of the importance of redundancy in speech for robust ASR.