Effects of the parameters of wavelets applied in de-noising of room impulse responses

Wavelets have commonly been applied in de-noising of various signals, especially speech and audio signals. However, to the best of knowledge of the authors, de-noising of room impulse responses (RIRs) by wavelets can hardly be found in literature. This paper presents the use of various wavelet algorithms for RIR de-noising. For this purpose, two types of synthesized RIRs are used. In order to have RIRs with different SNRs, synthetic white noise of different levels is added. The analysis is based on applying various wavelets, Daubechies, Haar, Coiflets, Symlet, Biorthogonal, Reverse biorthogonal and Mayer, either on room response to excitation or directly to extracted RIRs. The effects of wavelets' parameters, such as level and thresholding, are observed. RIRs obtained without and with wavelets are mutually compared in both time and frequency domain, and they are compared to the corresponding noiseless RIRs. Special attention is paid to the decay curves generated by backward integration of these RIRs. The results show that a significant improvement in dynamic range of the decay curve of even more than 20 dB can be achieved by applying wavelets. However, the technique is rather sensitive to wavelets' parameters so they should be optimally set.