Additive and multiplicative reestimation schemes for the sinusoid modeling of audio

This paper discusses an approach to accurate sinusoid modeling of audio. We propose an iterative framework which functions as a “wrapper” that works with arbitrary sinusoid modeling systems to boost their accuracy. It involves one or more reestimation steps. In each step the parameter estimates are updated by combining a second set of parameters evalulated from the latest modeling error signal. An additive scheme and a multiplicative scheme are proposed for this reestimation step. On a limited test set the framework is shown to offer 2dB to 40.5dB (average 14.6dB) improvement in signal-to-residue ratio within 5 updates, which is 56.3% to 98.6% (average 79.5%) of the largest possible improvement, in dB, obtained by interpolating exact parameters.