Modeling and coding of spot microphone signals for immersive audio based on the sinusoidal model

In this paper, the Sinusoids plus Noise Model (briefly SNM) is applied in a novel manner, in order to efficiently encode spot audio signals. These are the microphone recordings of a performance, before obtaining the multichannel mix, and are important for immersive audio applications since they can be used to provide interactivity. The SNM, as well as the SNM error spectral envelope, are extracted from each spot signal, providing a low-quality version of the signals. The main contribution of the paper corresponds to the use of a single audio reference signal which significantly enhances the quality of all the modeled spot signals. Reproduction of good quality and without loss of image width can be achieved using the proposed approach (above 4.0 perceptual grade for modeling and coding), by encoding a single audio (reference) signal, with side information per spot signal on the order of 19 kbps.