REMOVING CRACKLE FROM AN LP RECORD VIA WAVELET ANALYSIS
Pavel Rajmic, Jaroslav Klimek · 2004
The familiar “crackling” is one of the undesirable phenomena which we deal with in an LP record. Wavelet analysis brings a new alternative approach to the removal of this feature in the restoration process of the recording. In the paper, the principle of this method is described. A theoretical discussion of how the selection of the wavelet basis affects the quality of the restoration is also included. The wavelet-type signal analysis has recently been a much used discipline, whose range of applications in one-dimensional and multi-dimensional signal processing spreads steadily. It is used for the time-frequency analysis, for reconstruction of non-complete or strongly disturbed signals and for data compression in many fields. There were many algorithms developed how to restore the digitized audio signal from a vinyl record, for example SDROM (Signal Dependent Rank Order Mean) [3], methods based on linear prediction in AR and ARMA models [6] or the well-known median filtering. Another special class of restoration methods exploit the Bayesian statistics [5]. This paper introduces an alternative approach to the problem, using the wavelet signal processing. In the paper, we first discuss the time and frequency characteristics of a “crackle”. After this an overview of wavelet transform and its properties necessary for the method’s derivation is presented. Then, the principle of the crackle removal is described. At the conclusion, the discussion of how the choice of the so-called mother wavelet affects the quality and effectiveness of the restoration is introduced. 1.1. Characterization of a “crackle” The typical behavior of the crackle in the time domain usually corresponds to signal waveform “up and down”. We found that three quarters of these peaks stretch from 0.36 to 1.09 ms. The crackle has also bigger short-time energy than the rest of the signal. Another typical feature is that in spectral domain it spreads over all the frequency bands (see Figure 1). The character of the crackling also slightly differs depending on which part of the recording we work on ‐ this is due the different speed of vynil rotation at the beginning and at the end of the recording.