Noise estimation and detection algorithm should adopt to a changing environment in a fast manner so they use a LMS filter. However, there are some negative points as well. A LMS filter is very low and it consequently lowers a speech recognition rate. In order to overcome such weak point, I would like to propose a method for the establishment of a robust' speech recognition model in a noise environment. Since this proposed method allows the cancelation of noise with the AELMS filter in a noise environment, a robust speech recognition model can be established in a noise environment.