Accurate estimation of f(o) and aperiodicity based on periodicity detector residuals and deviations of phase derivatives

We introduce a new fundamental frequency (f(o)) and aperiodicity estimation method with its certified calibration data on estimation accuracy. The proposed method is based on two independent representations. The first representation is a set of band-wise residuals of a self-tuning process to suppress the prominent periodic component in the band-pass filtered output. The second representation is a set of band-wise deviation measure of the instantaneous frequency of the band-pass filtered output. The combined measure of these representations provides accurate estimates of SNR of each filtered band. This combination significantly reduces the distribution spread from 0.5% to 99.5% probability. The reduction is about 5 dB (for example, 21 dB to 16 dB). A wavelet-based configuration provides the initial estimate of f(o). A short term Fourier transform-based configuration provides refinement of f(o) and accurate estimate of aperiodicity. A MATLAB implementation of the procedure will be available as an open-source package.