Abstract
This paper describes the spectral modification method for signals polluted with noise. In this case, the spectral subtraction method is usually used. Instead of the spectral subtraction method, the spectral features of noisy signals are modified using some rules. However, irregular modifications occur in case the environment is polluted with colored noise. In this paper, improved modification rules for adding valleys to the spectral envelopes of signals under the noise environment is presented. Two types of threshold functions are presented and the stability of the intersection point between threshold functions and original spectral envelopes improved. The improvement of the spectral shape after this modification is shown and experimental results of pattern recognition are also shown.