Eliminating noise from a speech signal based on a pair of filters

Today, the increase in the number of devices working based on speech interfaces increases the importance of speech quality. However, even a minimal noise level can seriously affect the accuracy of speech signal recognition and processing. Therefore, denoising speech signals is an important task in signal processing, and it also serves to improve speech quality in telecommunications, speech recognition systems, and other speech-related applications. Applying existing noise reduction filters separately may not always be effective. Therefore, in this research, a noise reduction approach based on the sequential application of filters is proposed. Based on literature analysis, filters such as low-pass, band-pass, Kalman, Butterworth, and elliptic filters were selected, and pairs were formed based on them. Pairs of filters were applied to speech signals with different levels of noise, and the resulting filtered speech signals were evaluated based on the PESQ evaluation criterion. The purpose of this study is to determine the optimal pair of filters that can minimize the impact of noise on the quality of speech signals using the PESQ evaluation criterion. The results of the experiments showed that it is optimal to use a pair of band-pass and Butterworth filters at a low level of noise, a pair of low-pass and elliptic filters at a medium level, and a pair of band-pass and elliptic filters at a high level of noise. The results obtained are important and practical in the development of other hybrid methods of noise reduction.