This paper presents a joint estimation approach to extract multiple fundamental frequency (F0) in monaural polyphonic music signals. In a frame-based analysis, we generate a spectral envelope for each combination of F0 candidates, from non-overlapped partials, under assumption that a harmonic sound is characterized by a Gaussian mixture model (GMM). The optimal F0 candidates combination minimizes a spectral Euclidean distance measure between the original spectrum and Gaussian spectral models. Evaluation was carried out using several piano recordings. Evaluation show promising results.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.