Sinusoidal interpolation across missing data

We introduce a method for the concealment of a missing segment in an audio signal. The method is based on sinusoidal interpolation and extrapolation operations. An extrapolation is always followed by an interpolation and an interpolation is always followed by observed data. For interpolation, sinusoids are detected in both the observations prior and after the missing data segments. All sinusoids together are sorted by their amplitude before pairing up sinusoids observed before and after the missing segment. The complex amplitude and the frequency of the sinusoids are linearly interpolated. For extrapolation constant amplitudes and frequencies are used. The interpolation method can operate in two modes: a synchronous mode and an asynchronous mode. In the synchronous mode, the length of the segment of missing data is fixed and assumed to be consistent with the model. In the asynchronous mode, the length of the missing data segment is selected to minimize amplitude excursions over the interpolation interval. Experiments confirm that the new error concealment method performs significantly better than existing packet loss concealment methods for music signals.