SUBBAND-BASED PARAMETER OPTIMIZATION IN NOISE REDUCTION SCHEMES BY MEANS OF OBJECTIVE PERCEPTUAL QUALITY MEASURES
Thomas Rohdenburg, Volker Hohmann, Birger Kollmeier · 2006
In general, noise reduction schemes for application in hearing-aids or car environments have parameters that are determined by technical distance measures or heuristically based on infor-mal listening by the algorithm developers. In [1] we have shown that quality measures based on psychoacoustic models are better suited to optimize single parameters in terms of the best subjec-tive overall quality than pure technical measures like, e.g., the signal-to-noise ratio. In other words, a test-bench based on ob-jective quality measures and several typical noise types can sup-port the search for the best-sounding noise reduction algorithms and their internal parameter settings. However, if the algorithms becomemore complex, e.g., because of frequency-dependent pa-rameters, a single broadband measure might not be feasible to assess optimal settings because of the high dimensionality of the parameter space. In this case a subband-based perceptual quality measure might be feasible. In this study, we exemplarily apply subband-based quality prediction to parameter optimization in a noise reduction algorithm based on auditory filters. 1.