Spectral similarity metrics for sound source formation based on the common variation cue

Mathieu Lagrange (1) , Martin Raspaud (2)

1 IRCCyN - Institut de Recherche en Communications et en Cybernétique de Nantes
2 LaBRI - Laboratoire Bordelais de Recherche en Informatique

Scene analysis is a relevant way of gathering informations about the structure of an audio stream. For content extraction purposes, it also provides prior knowledge that can be taken into account in order to provide more robust results for standard classification approaches. In order to perform such scene analysis, we believe that the notion of temporality is important. We study in this paper a new way of modeling the evolution over time of the frequency and amplitude parameters of spectral components. We evaluate the benefits of such an approach by considering its ability to automatically gather the components of the same sound source. The evaluation of the proposed metric shows that it achieves good performance and take better account of micro-modulations.

Domaines: Machine Learning [stat.ML]

Mathieu Lagrange, Martin Raspaud. Spectral similarity metrics for sound source formation based on the common variation cue. Multimedia Tools and Applications, 2010, pp.185-205. ⟨hal-01132571⟩