Application of the fan-chirp transform to hybrid sinusoidal+noise modeling of polyphonic audio

Reliable classification of spectral peaks as tonal and noise-related is an important stage of hybrid sinusoidal+noise modeling. Spectral peaks of higher harmonics are often missed due to their wide frequency spread resulting from pitch variation. Recently introduced fan-chirp transform allows for compensating the changes of fundamental frequency in the process of spectral analysis of speech and harmonic sounds. In case of polyphonic audio the fundamental is often not unique and/or is hard to estimate. We propose a simple technique for estimation of chirp rates from already detected partials to improve the detection of higher harmonics through the application of frequency warping and fan-chirp analysis.