Beta 1

Title Modeling the sharpening of tuning in nonsimultaneous masking
Author Marschall, Márton
Supervisor Dau, Torsten (Department of Electrical Engineering, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Buchholz, Jörg (Department of Electrical Engineering, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2008
Abstract Frequency selectivity refers to the ability of the auditory system to resolve sounds in the spectral domain. In humans, frequency selectivity can be measured behaviorally with masking experiments. Results from such experiments show greater frequency selectivity when nonsimultaneous masking techniques are used as opposed to simultaneous ones (Moore et al., 1984; Oxenham and Shera, 2003). The differences in tuning have been suggested to occur because of suppression, a nonlinear interaction between the masker and the signal, whose effects are only observed when the two stimuli overlap in time. Suppression is thought to be a consequence of the compressive behavior of the basilar membrane. As a result of suppression, the increased frequency selectivity in nonsimultaneous masking may enhance the perception of dynamic signals, such as speech. In this thesis, the relationships between frequency selectivity, compression, and suppression were studied using a modeling approach. A computational auditory signal processing model developed by Jepsen et al. (2008) was used to simulate two behavioral measures of frequency selectivity: psychophysical tuning curves and the notched-noise method. The model included the dual-resonance nonlinear (DRNL) filterbank (Lopez-Poveda and Meddis, 2001). Simulations were compared with the results of a pilot test on one test subject, and with published data. It was found that changes were needed to the parameters of the DRNL in order to account for the tuning differences observed in human data. Specifically, from the modifications suggested by Plack et al. (2002), the most important one was the use of different bandwidths and center frequencies for the bandpass filters in the nonlinear path of the DRNL. Simulations with the modified model were in good agreement with human frequency-selectivity estimates, although discrepancies were observed with respect to the threshold differences between simultaneous and forward masking. As a result of the modifications, an auditory signal processing model was obtained that shows instantaneous changes in frequency selectivity due to suppression effects. In further studies, such a model could be used to investigate how changes in tuning affect the perception of speech-like dynamic signals.
Pages 67
Original PDF marschall_thesis.pdf (0.99 MB)
Admin Creation date: 2009-05-06    Update date: 2009-05-06    Source: dtu    ID: 242420    Original MXD