DETECTION OF CLIPPED FRAGMENTS IN ACOUSTIC SIGNALS

Abstract

The paper deals with investigation of the method for detecting clipped fragments in acoustic signals with better characteristics as compared with the other known methods. This method is based on the histogram construction for the analyzed signal amplitudes and calculating the distances between the local peaks of the histogram on its tails and in the central part. The difference between histograms of non-clipped and clipped signals is that the histogram of a non-clipped signal has smoothly decaying tails while the histogram of a clipped signal has visible and easily detectable outbursts on its tails. The value of these outbursts and consequently the quality of detection of clipped fragments depends on the parameters of the method under investigation. The main aim of this paper is finding the optimal parameters of the method. Characteristics of the method are studied in detail by mathematical modeling; density functions of target values for different lengths of a studied signal frame and the number of histogram counts and levels of clipping of acoustic signals are built. It is shown that good separation between clipped and non-clipped signal fragments of acoustic signals can be achieved for the frame length between 6000 and 8000 samples and the number of histogram bins between 200 and 300. In this case the threshold level for the best separation can vary between 0.45–0.55. Examples of clipping detector operation based on the proposed method and on real acoustic signals are shown for the case of different clipping levels

Keywords

• acoustic signal
• clipping
• clipping coefficient