An Approach for Noise-Speech Discrimination Using Wavelet Domain

Abstract

There are many sudden and short period noises in natural envrioments. In this paper, noise reduction is efficiently performed for additive white noise and an automatic thresholding method for discriminating of noise / speech. This modified version of the thresholding method updates the threshold in each frame. In this proposed method, the selection of the threshold value depends on the estimates of the standard deviation and gives it as the input to the super-soft thresholding algorithm. Voice activity detection methods usually work in time or frequency domains. We propose super soft thresholding algorithm based on subband voice activity detection. If clean speech data can be input, it will help prevent system operations errors. These proposed methods are applied in a real-time noise reduction.