Mitigation of long gaps in music using hybrid sinusoidal+noise model with context adaptation

The paper describes a technique allowing to conceal or mitigate long gaps (up to 0.5 second) in music programs. Missing content is replaced by signal synthesized from spectral models using data surrounding the gap. Tonal components are synthesized using a sinusoidal model. A heuristic adaptive algorithm is employed to link model parameters across the gap. Prior to linking, sinusoidal partials are categorized as stable or variable, allowing to properly dealing with vibrato or glissando notes in music. The noise part is synthesized using a warped LPC model. Results of blind listening tests are reported.