A comparative study on AM and FM features

In this paper, we investigate the advantages of frequency modulation (FM) features by conducting speech recognition experiments and statistical analysis. The importance of temporal aspects in speech recognition has been discussed along with the importance of amplitude modulation (AM) and frequency modulation. Recently, we have proposed a speech recognition system that is based on the combination of AM and FM features and confirmed its efficiency experimentally. Although the proposed speech recognizer assumes complementarity between the AM and FM features, it was not evaluated in previous studies. In this paper, in order to evaluate the complementarity between two types of features, we conducted continuous digit recognition tasks in artificial noisy conditions. We confirmed that the error rates of each classifier are significantly different depending to kind of noise. Then, we evaluated the statistical independency between these two types of features. We confirmed that the behaviors of these features are independent in realistic noisy environments.