Noise Reduction Using Thresholding-Based Algorithm and Homogeneous-Frame-Merged Approach for Wavelet-Packet Transform

Improving speech quality for a corrupted signal is still an important task for voice communication. This investigation proposes to incorporate thresholding-based algorithm with homogeneous-frame-merged (HFM) approach to adapt a gain factor. In turn, we employ this gain factor to reduce the noise components for a corrupted signal. Firstly, a fixed-length frame is employed to calculate the power of noisy wavelet coefficient. Afterwards, merging homogeneous frames in a segment is performed to estimate the power of speech wavelet coefficient. Substituting this speech power in wavelet coefficient to the thresholding-based algorithm, enhanced speech signal is then obtained. Experimental results show that the proposed gain factors vary smoothly among homogeneous frames for each subband, enabling the residual noise to be reduced significantly.