Sound representation and modification with multiresolution sinusoidal models
Michael M. Goodwin, Paolo Pradoni, Martin Vetterli · The Journal of the Acoustical Society of America · 1998
In the standard sinusoidal model, the analysis and synthesis are typically carried out at fixed frame rates. This can result in representation error for signals with dynamic behavior. To improve the modeling performance for such signals, the sinusoidal analysis-synthesis can be carried out in a multiresolution framework based either on a filter bank structure or on adaptive time segmentation. In the latter method, which will be of primary interest here, the adaptation is guided by a metric such as rate-distortion. Such an approach improves the model accuracy while preserving the ability to carry out the rich class of modifications afforded by the standard sinusoidal model.