Modifications by signal Sample Clauses

Modifications by signal processing Alterations by means of signal processing form another part of the editing history of an audio signal. Detection techniques typically aim at operations that occur at specific places in this history, including content production, the application of audio effects, or nonlinear processing that might be induced by digital forgeries. Estimating mixing parameters The parameters used in the production of audio contents, in particular in the processes of mixing and mastering, provide insight into the history of an audio track. In [201], a method to estimate these parameters is proposed. Both linear parameters (mixing gains or linear time-invariant filters) and nonlinear processes are targeted by this technique. The supported nonlinear elements are modeled as a gain controlled by a time-varying envelope that is calculated from a level measurement of the signal. This model includes a wide range of common audio effects such as limiters, compressors, or expanders. The mixing parameters are determined by using a least-squares technique. A basic assumption of this approach is that the input tracks are linearly independent. It is shown that this property holds for most audio signals, even if source signals interfered during recording. While the determination of the mixing parameters is not unique in many cases, i.e., if two gains in a processing chain can be adjusted to yield the same final gain, it is shown that the least-squares algorithm achieves a non- degenerate solution in such cases. While linear parameters such as gains or filter coefficients can be calculated for the complete audio signal, the gain envelope corresponding to a nonlinear processing element is calculated frame-wise. For each audio frame, this envelope is modeled by a polynomial whose coefficients are determined using a least-squares criterion. However, the proposed parameter estimation requires the input tracks of the mixing process to be present. This is a major drawback for the use in footprint detection, which is typically considered as a blind process. In addition, the size of the least-squares problem grows with the number of estimated parameters. As an example, if filters are modeled in the mixing process, the size of involved matrices grows linear with the assumed filter order. Besides the increased computational complexity, such problems are prone to ill-conditioning. A high sensitivity to noise is noted by the authors as a main disadvantage of the algorithm. It appears likely th...