Time-differential encoding of sinusoidal model parameters for multiple successive segments

Sinusoidal coding is a key technique for low rate audio coding. In sinusoidal coding, the target signal is represented by perceptually relevant sinusoids; however, often the sinusoids are estimated without taking into account that the sinusoidal parameters are going to be differentially encoded. In this paper we present an algorithm for joint extraction and time-differential encoding of sinusoidal model parameters. For a pre-specified target bit rate, the algorithm extracts the set of sinusoids for a sequence of signal segments which lead to minimum distortion in the reconstructed signal. Furthermore, it determines which sinusoids in a given segment should be encoded time-differentially relative to which in the previous segment. Simulation experiments show that the proposed algorithm leads to a reduction of 3-5% in bit rate compared to a state-of-the-art time-differential sinusoidal coding system.