Spectral Modulation Sensitivity Based Perceptual Acoustic Echo Cancellation

Abstract

Perceptual acoustic echo cancellers were developed by mainly considering human hearing thresholds of different acoustic frequencies in the past. In addition to the different frequency sensitivities, the human brain further analyzes sounds in terms of their spectral and temporal modulations. In this paper, we extend the perceptual normalized least mean square (P-NLMS) algorithm by adding a second pre-emphasis filter which accounts for brain's spectral modulation sensitivity. Simulation results and listening tests demonstrate our system effectively reduces the perceived residual echo during convergence due to faster convergence rates in hearing-sensitive spectral modulation bands and acoustic frequency bands.