We revisit the problem of blocking artifacts and their suppression in generic frame-based speech/audio applications. We provide a perceptual characterization of the artifacts by using dynamic auditory models. We propose some short-time-Fourier-transform-based magnitude and phase smoothing techniques and show that localized time-frequency smoothing suppresses the artifacts to a large extent. Our experiments show that magnitude smoothing is superior to phase smoothing and that the latter turns out to be only detrimental to the signal quality. We provide some examples on natural speech and audio signals in the context of compression. (C) 2008 Elsevier B.V. All rights reserved.