Gaussian Framework for Interference Reduction in Live Recordings
Résumé
In this study, typical live full-length music recordings are considered. In this scenarios,
some instrumental voices are captured by microphones intended to other voices, leading
to so-called “interferences”. Reducing this phenomenon is desirable because it opens new
possibilities for sound engineers and also it has been proven that it increase performances
of music analysis and processing tools (e.g. pitch tracking). Extending state-of-the-art
methods, we propose an NMF-based algorithm that iteratively estimate each source contribution,
i.e. the power spectral densities (PSDs), and the corresponding strength in each
microphone signal, modeled in a interference matrix. Unfortunately our approach suffer
of a huge computational load. To address this issue, We show that using random projection
method the method is able to process full-length live multi-track recoding in a
acceptable time. Experimental results demonstrate the efficiency and the effectiveness of
the approach.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...