Real-time audio error concealment method based on sinusoidal model

This paper proposes a sinusoidal model based on modified sinusoidal analysis and synthesis for the error concealment of lost frames. The parameters of the sinusoidal model are adjusted according to the property of the audio signals. It is applied to MPEG-2 AAC decoder for an error concealment of decoder based only. An analysis and synthesis of the windowed previous frame of the lost one and the residual samples are used to synthesize a new frame. Then overlap-add mechanism between the synthesized frame and the residual samples is applied to reconstruct the lost frame. This method can provide an audio quality with 1.3 average ODG score improvement under the loss rate of frame up to 10% without increasing the algorithm delay of the AAC decoding system. It is implemented on fixed-point DSPs, and the computational complexity is comparative to the optimized version of the decoder.