Sound restoration in damaged acoustical recording

Modeling of an acoustical signal in transformed form is presented. The signal is transformed to the frequency domain, and the behavior of each spectral component is modeled. Estimation of each spectral component for a time frame is made from the individual models. The spectral frame is then transformed into the corresponding time frame. The application under consideration is restoration of sound in old acoustical recordings which have lost some portion of the recording due to either deterioration or heavy noise. The attractive feature of this method, in addition to getting good results, is that it uses commonly available hardware like PC AT with sufficient RAM. The computations needed in modeling involve matrices at lower rank, and hence the error is minimal. The processing is done digitally through FFT and IFFT. Overlapping of input data is done in order to minimize distortion of the signal due to truncation