Comments on "Towards increasing speech recognition error rates" by H. Bourlard, H. Hermansky, and N. Morgan