An initiative for testability and comparability of binaural models

Dietz M, Marquardt T, Majdak P, Stern RM, Hartmann WM, Goodman DF, Ewert SD
Acoustical Society of America (2017)
doi: 10.1121/1.4987810
Acoustical Society of America meeting 2017
 
Abstract. A framework aimed at improving the testability and comparability of binaural models will be presented. The framework consists of two key elements: (1) a repository of testing software that evaluates the models against published data and (2) a model repository. While the framework is also intended for physiological data, the planned initial contribution will be psychoacoustical data together with their psychoacoustical testing protocols, as well as existing binaural models from available auditory toolboxes. Researchers will be invited to provide their established as well as newly developed models in whatever programming language they prefer, given the models are compatibility with the proposed interface to the testing software. This entails that the models act as artificial observers, testable with exactly the same procedure as the human subjects. A simple communication protocol based on wav and txt-files is proposed because these are supported by every programming environment, and are able connect models and testing software of any programming language. Examples will illustrate the principle of testing models with unaltered signal processing stages on various seminal data sets such as tone detection in so-called double-delayed masking noise, or lateralization of ¾-period delayed noise and sounds with temporally asymmetric envelopes.
 
Links: Conference
 
Categories: Auditory, Neuroinformatics, Neuroscience, Sensory, Software