Primena talasića na otklanjanje šuma i parametrizaciju audio signala

Methods for signals analysis have been developing more and more in last few decades. By discovering new ones or by upgrading the existing algorithms, results of signal processing have been significantly improved, especially if the emphasis is on the process of removing noise from the signal or signal parameterization. A number of analog and digital filters are present in many scientific fields nowadays, and a technique that is more and more in use is the one based on wavelets. Wavelet decomposition to its detail and approximation coefficients is the method that provides significant results in removing noise from the signal as well as in signal parameterization. This thesis presents the results of applying wavelets for the reduction of noise and its negative effects in synthetized and measured room impulse responses. Analysis is done with signals in full frequency range (broadband signals) and signals filtered in octave and third-octave bands. Different sets of wavelet parameters are analyzed with the aim of finding the optimal set of parameters that will give the best results. White and pink noises are used in the whole research observing their influences on room impulse responses and energy decay curves generated from these responses. Two used measures that quantify the achieved results are the dynamic range improvement of the decay curves and noise floor reduction. Other standard methods for noise removal and reducing noise effects in room impulse responses are used for the purpose of comparison with results achieved with wavelets. Besides the method of noise removal, wavelets have also found their place in the knee detection in energy decay curves obtained from room impulse responses. The obtained results are also compared with the standard methods from literature. An important task when it comes to newly manufactured industrial products, such as direct current motors, is to find procedures and algorithms that will separate non-faulty and faulty motors. Method that can be used for this purpose is related to application of wavelets to audio signals containing sound generated by analyzed motors. The detail coefficients obtained by the wavelet decomposition process representing wavelet-based audio features may show the mentioned differences between motors at certain levels of decomposition. The achieved results are confirmed by applying the introduced numerical measure called feature difference of the detail coefficients at certain levels