Audio declipping using sparsity-based methods
Audio signals are often disposed to different types of quality degradation. One of the most common degradations is amplitude clipping (saturation), which is related to the dynamic range limitation. Clipping is not only perceptually unpleasant, but it also reduces the accuracy in applications such as automatic speech recognition, voice-based Parkinson’s disease detection, etc. The presentation will be devoted to the current state-of-the-art approaches to audio declipping, which are in majority based on the signal sparsity. The lecture will show how the audio restoration tasks can be formulated and what algorithms can be used to compute the solution. It will discuss the properties of the state-of-the-art algorithms, their computational complexity, and it will also tackle related problems such as incorporating psychoacoustics into the restoration process or enhancing the results of declipping methods that allow a deviation in the reliable part by postprocessing methods. Finally, the lecture will explain the evaluation of the restoration quality and mention possible directions for future research in the field.