High-frequency Noise Removal of Audio Files using Daubechies Wavelet Transform

Abstract

In general, audio signals are contaminated with various types of noise. This paper presents a novel signal processing method developed for high-frequency noise elimination using wavelet transforms. As a continuation of a previous study that used Fourier transform for noise removal in audio files, in this study Daubechies wavelets were used to reduce computational complexity and achieve better noise reduction performances. Compared to the Fourier transform, the Daubechies wavelet transform method removes the noise in each signal while preserving its vital characteristics. The suitable level of the Daubechies wavelet for noise removal in each channel was obtained using a trial-and-error approach. It was identified that the ideal range for the level of the Daubechies wavelets for noise removal is between 17 and 20. Moreover, unlike the Fourier transform, the Daubechies wavelet transform demonstrates a proficient capacity in eliminating noise from data point that lies completely outside the rest in the audio data set. Wolfram Mathematica 12.3 software package was used to complete this research. This method can be applied to
conserve vintage audio recordings originally recorded in cassettes and spools.

Keywords: Digital Signal Processing, Wavelet Transforms, Daubechies Wavelet Transform, Fourier Transforms, Noise Removal