Exploration of the Additivity Approximation for Spectral Magnitudes

Stephen D. Voran

Abstract: The separation of acoustic signals is often accomplished through subtractive decompositions of frequency-domain representations. This is typically enabled by the zero phase approximation or the uncorrelated signals approximation but both of these are very coarse approximations in the mathematical sense. We investigate this disconnect between what works in practice and what is mathematically correct. We conduct a broad search for a domain where the additivity of spectral magnitudes is best satisfied. We apply objective estimators to time-domain reconstructions to characterize the true auditory impact of the magnitude additivity approximation. Our results show the auditory impacts of additivity approximations and allow comparison with the impact of using mixture phase and exact magnitudes in the time-domain reconstruction.

Keywords: noise reduction; speech enhancement; auditory scene analysis; compositional model; noise suppression; source separation; spectral subtraction

(Voran_WASPAA15.pdf)

Disclaimer: Certain commercial equipment, components, and software may be identified in this report to specify adequately the technical aspects of the reported results. In no case does such identification imply recommendation or endorsement by the National Telecommunications and Information Administration, nor does it imply that the equipment or software identified is necessarily the best available for the particular application or uses.

Back to Search Results