Pavel Rajmic

This paper revisits the autoregressive (AR) interpolation method proposed by Etter, and provides an open implementation and proposes two practical extensions: A formulation that allows high-order AR models even when the gap length is shorter than the model order, and a causal variant tailored to packet loss concealment.

LOUDAR adapts to unknown audio distortions on-the-fly, achieving superior restoration results without the need for paired training data.