Published Paper Details:

The goal of this research is to see how well the wavelet transform performs while de-noising a spoken signal. Wavelets are frequently employed in digital voice processing, particularly in speech signal coding, augmentation, and noise removal. Because of the background noise of genuine speech, it might be difficult to recognize it in many situations. A voice denoising algorithm's purpose is to recover the original speech signal by reducing noise with the least amount of distortion possible. There are a number of techniques that may be used to assist recover speech that has been distorted by noise. Many of the commonly used denoising techniques do so in the frequency domain, where the noisy signal's power spectral density (PSD) function can be analyzed in a short time period. Then, for each frame of the noisy data, the short-time spectral frequency and amplitude of clean speech are computed. As a result of the limits of methodologies, estimate mistakes are introduced. For decades, many spectrum estimating strategies have been studied in order to decrease estimation mistakes. The discrete wavelet transform technique is employed in this work to denoise an input noisy voice stream. Different wavelet filters, such as Daubechies, Symlets, and Coiflets, are used to evaluate the performance of discrete wavelet transforms. MATLAB software was used to do the analysis. Different sorts of environmental background noises were studied as input noisy speech signals, such as babbling noise (crowd of people) or noisy talks with various types of background vehicle noises (cars, train, plane etc.). With hard or soft thresholding approaches, the input noisy speech signal was decomposed by applying four alternative threshold selections to the wavelet coefficient: sqtwolog, heursure, rigrsure,and minimaxi thresholding. The signal-to-noise ratio (SNR) and MSE values between noisy and output signals were used to compare reconstructed speech to the original speech signal. Detailed comparisons of several wavelet families' capabilities against various background noise types are included, as well as the discovery of an efficient approach (Maximal overlap DWT-MODWT) for denoising noisy voice signals.