Restoration of Speech Signals Contaminated by Stationary Tones Using an Image Perspective

Instead of using the traditional approach of attenuating the speech frequencies which are contaminated by tones, the project uses image processing techniques to do an intelligent restoration of the contaminated frequencies. The first step is to obtain the fundamental frequency patterns which are repeated periodically throughout the image spectrogram. These patterns are obtained by using cepstral estimation. The following step is to restore the frequencies affected by the stationary tones making use of the shape and location of frequency patterns found in the previous step. The concept of Markov random fields (MRF) used in estimation problems of image processing is used for this restoration phase