Speech denoising using discrete wavelet packet decomposition technique

Abstract

In this paper, a discrete wavelet packet transform algorithm is used for speech signal denoising. Both hard and soft thresholding are applied and noisy speech signal samples corrupted by white Gaussian noise from 0dB to +15dB are denoised in our experiments. Output SNR (Signal to Noise Ratio) values are calculated and compared with input SNR values using both types of thresholding methods. Soft thresholding method is observed to perform better than hard thresholding at all input SNR levels. Hard thresholding shows a maximum of 5.9 dB improvement whereas soft thresholding shows a maximum of 6.5 dB improvement as an output SNR value.