This paper presents a test normalization (TN) technique to improve the robustness of an Automatic Speech Recognition (ASR) system. Earlier experimentations of automatic speech recognition show that the application of sub-band spectral subtraction (SSS) is very useful for the reduction of the effect of noise from speech at signal level to improve the recognition accuracy rate in different testing and training conditions. In this work, at the testing process, the estimated scores of each speech signal are normalized with a test normalization technique. Hidden Markov Model (HMM) is used for training and testing process of the ASR system. The ASR experimentation result shows a little improvement of robustness with the combination of SSS and TN approach.
