Comparison of noise reduction filters for speech enhancement

Abstract

Speech is one of the best and easy ways for communication. In speech communication, the speech signal is affected by some background noises, additive white Gaussian noise (AWGN) as well as distortions from Rayleigh and Rician channel. The performance of speech processing is reduced by background noise which makes it difficult to listen to the speech and is the major reason for speech quality degradation. Noise reduction techniques play a key role in removing the artifacts. These noise reduction techniques are used for separating the clean speech from noisy observations. In this thesis, two algorithms are used to eliminate the noises in single channel speech enhancement such as; spectral subtraction and Wiener filter. The main motive of these algorithms is to reduce the level of noise and improve the corresponding signal-to-noise ratio (SNR) at different power levels of input SNR. This paper involves the theoretical and practical analysis of these algorithms. In which wiener filter is compared with the spectral subtraction algorithm based on SNR improvement introduced by them. These noise reduction techniques show that output SNR is more than the input SNR. This means, filters are able to remove the noise from noisy speech. In whole, experimental results show that SNR improvement of the Wiener filter is much better than the implementation of spectral subtraction.