A Wavelet-Based De-Noising Speech Signal Performance with Objective Measures

Abstract

AbstractSpeech signals are complex. Their quality can diminish drastically due to the interference of different types of noises. A system is required to de-noise the noisy speech signals and turn them into clean speech signals. As these types of noises are large, enhancing speech signals often becomes a difficult task. Traditional speech enhancement algorithms like spectral subtraction and Weiner filtering prove unsatisfactory in a non-stationary noise environment. Thus, removing noise and getting a clean speech directly are difficult, so we need immense methods to tackle the noise. In this study, we use discrete wavelet transform and MATLAB software to de-noise a speech signal taken from the speech corpus database. The objective of our proposed method is to reduce the noise of a speech signal better than the previous methods. This method deals with the speech signals of various areas where there is more noise which is difficult to get plain speech signals. The threshold values of the noisy signal are taken down first and then are compared with the de-noised signal. The performance of the proposed method is analyzed by objective speech quality measures like signal-to-noise ratio, segmental snr and frequency spectral SNR. The results show that the proposed model with the DWT feature improves the quality and intelligibility of a speech signal. The proposed model is easy to implement and helps to reduce the noise of a speech signal.KeywordsTraditional speech enhancement algorithmsSpectral subtractionWeiner filteringNon-stationaryDiscrete wavelet transformSpeech corpusThreshold valuesSignal-to-noise ratioSegmental SNRFrequency spectral SNR