Development of a Blackman Window-Based Soft Thresholding Function for Wavelet Denoising of Speech Signals Corrupted by Additive White Gaussian Noise

Full abstract

Read the full abstract

The most significant and efficient way for humans to communicate is through speech or the human voice. If the integrity of the message being conveyed by the transmission lines must be preserved at the receiving end those noise components must be removed so as to enhance the signal quality. In this research, the objective is to denoise a voice or speech signal affected by the most common noise component that interferes with it, which is additive white Gaussian noise (AWGN), using discrete wavelet transform. A new shrinkage function, called the Blackman window-based soft thresholding function, is developed utilizing the Blackman window. Based on the developed shrinkage function, with a decomposition level of 2 and using the Symlet 4 mother wavelet, the denoising of the contaminated voice/speech signal was extensively performed using four different threshold estimation rules, including sqtwolog, rigrsure, heursure, and minimaxi. The results obtained were a power spectral density of -68.23 dB at 0.5 radian normalized frequency, signal-to-noise ratio of 25.553 dB, mean square error of 0.0108, and maximum absolute error of 0.07870, indicating that the model was effective. The information provided by this study will enhance the Blackman-based, window-based shrinkage soft thresholding function as a reliable alternative to other well-known wavelet coefficient shrinkage functions in wavelet-based techniques for practical applications.