|
|
Modern Optimization Methods for Interpolation of Missing Sections in Audio Signals
Mokrý, Ondřej ; Kowalski, Matthieu (oponent) ; Koldovský, Zbyněk (oponent) ; Rajmic, Pavel (vedoucí práce)
Damage to audio signals is in practice common, yet undesirable. Information loss can occur due to improper recording (low sample rate or dynamic range), transmission error (sample dropout), media damage, or because of noise. The removal of such disturbances is possible using inverse problems. Specifically, this work focuses on the situation where sections of an audio signal of length in the order of tens of milliseconds are completely lost, and the goal is to interpolate the missing samples based on the unimpaired context and a suitable signal model. The first part of the dissertation is devoted to convex and non-convex optimization methods, which are designed to find a solution to the interpolation problem based on the assumption of sparsity of the time-frequency spectrum. The general background and some algorithms are taken from the literature and adapted to the interpolation problem, many modifications and experimental approaches are original. The second part of the thesis focuses on the use of non-negative matrix factorization, with which a probabilistic model of the signal spectrogram can be constructed and used for the interpolation of the signal. This model is then used as the basis for a successful reconstruction algorithm, to which two alternative methods are derived in the present thesis. Finally, an extensive experimental validation of the methods on a group of musical signals is conducted. Using objective indicators of the quality of the interpolated signal, it is shown, that in each class of methods, the proposed modifications lead to a noticeable improvement in quality or convergence over the baseline methods. In particular, within the studied range of impairments, algorithms using factorization compete with the current best methods for interpolating missing sections of the audio signal.
|
host ::
RSS.