Elimination of unwanted signals in audio materials using wavelet transform

As a sound engineer, employed at Croatian Radio, I realized that the tools for the reconstruction and restoration of old recordings (or damaged recordings) are insufficient and at least they have deficient analyzing method. Orthogonal frequency-based systems, such as the DFT do not offer a good insight into the temporal localization of unwanted parts of the audio material such as pulse-formed signal (e.g., clicks). In this search for such a tool, I reached for the discrete wavelet transformation (DWT) which is realized by double decomposition in order to obtain wavelet coefficients and a graphic depiction of the coefficients. Distortion is measured with according mean square error (MSE) and it is compared with the number of discarded wavelet coefficients. It is also done the comparison between DWT and the results obtained with frequency based systems based on discrete Fourier transform (DFT). After being performed on an arbitrary mathematical function, the wavelet transform is applied on a sound example. (Wave 22.05 kHz, 8 bit). It is shown that DWT is dealing very well with both, noise and discrete disturbance which are the most common problems in the daily work with audio material.