Statistical Evaluation of Localization Performance and its Application in an Internet-Based Self-Screening Test Using a Virtual Environment

This paper presents a method for scoring forced choice hearing tests with more than two alternatives. The proposed scoring procedure classifies the observed sequence of choices in a maximum likelihood sense. The classifier exploits a priory knowledge about the choice statistics estimated from a limited set of observed data. The proposed method was implemented and evaluated in conjunction with a newly developed Internet-based self-screening test. The implemented test measures localization skills in an interactive auditory virtual environment (AVE) while including both acoustical and behavioral aspects of everyday localization situations. By means of cross-validation, it is shown that the test results of listeners with normal localization skills and the test results of listeners with impaired localization skills can be reliably separated using the introduced scoring method. It is further shown that good classification results are obtained when a uniform a priori density model for the choices of the localization-impaired reference group is applied.