A method for evaluation of parasitic frequency modulation (wow) in archival audio is presented in this paper, published in the PrestoSPACE project. The proposed approach utilizes sinusoidal components as their variations are highly correlated with the distortion variations. The sinusoidal components are extracted from audio signal by means of sinusoidal modeling procedures being often severely distorted and in case of wow also signficantly modulated.
The algorithm for sinusoidal component evaluation utilizes both magnitude and phase spectra information to enhance the tracking process. Additionally, a neural-network based prediction module is proposed to improve the tracking abilities in case of component discontinuities. Experiments concerning prediction of tonal component’s values are performed revealing that prediction can enhance sinusoidal modeling of wow distorted signals effectively.