Extraction of spectral features of signals under the noisy environment using modification rules and weighting function

This Paper describes the method for effective extraction of spectral features of signals under the noisy environment. Further examination showed that extraction of the features of signals under the noisy environment is possible using modification of spectral envelopes by rule. In this modification method, vanished valleys of spectral envelopes under the noisy environment are recovered. However, the incorrect valleys are sometimes added and the obtained envelopes have bad influences on performance of systems using these features. In this paper, smoothing functions are used for avoiding wrong spectral valleys by rule and the improvement of performance for extraction of spectral features is shown. Four kinds of smoothing functions are defined and evaluated. Experimental results of pattern recognition using proposed spectral parameters are presented.