Real time adaptive denoising of musical signals in wavelet domain

In this paper, an adaptive scheme of real time denoising of musical signals corrupted by white Gaussian noise based on wavelet transform is presented. The proposed scheme is adaptive in nature and does not require manual tuning of any parameter. It automatically estimates the amount of noise present in the signal and calculates the required value of threshold for each level of wavelet decomposition. Therefore, proposed scheme is able to handle different types of musical signals irrespective of their types and length. The scheme is implemented on Texas Instrument´s digital signal processor TMS320C6713 in the MATLAB development environment and tested for various types of inputs. The effect of frame size and types of wavelets on the performance is also investigated. The proposed model shows good results both in terms of PSNR and audible quality of signal.