Audio coding using steady state harmonics and residuals

In this paper, we solve the problem of inter and intra-frame frequency leakage in harmonic based audio coding approaches. Specifically, we first use an FFT to estimate the harmonic parameters. Because of the limited frequency resolution of the windowed FFT, there is an uncertainty in the measured frequency within one frequency bin. To minimize this uncertainty, we use these estimates as seeds to a minimum least squares problem and solve for the optimal harmonic parameters. The resulting residual should be essentially free of harmonics and consist of only attacks and noise. We are preserving the attacks and coding them separately. The resulting residual noise is modeled and coded using wavelet transforms and noise modeling. We also present a solution to the problem of inter-frame frequency fading and highlight the benefit of using a rectangular coordinate representation of the harmonic data. The improved coder that we propose can be viewed as an object based approach to audio coding and leads to perceptually pleasant progressive transmission and representation schemes for multimedia applications