Frequency selectivity in the human auditory system is often measured using simultaneous masking of tones presented in notched noise. Based on such masking data, the equivalent rectangular bandwidth (ERB) of the auditory filters can be derived by applying the power spectrum model of masking and assuming a rounded-exponential filter shape. If a forward masking paradigm is used instead of simultaneous masking, filter estimates typically show significantly sharper tuning. This difference in frequency selectivity has commonly been related to spectral suppression mechanisms observed in the cochlea. Considering bandwidth estimates from previous studies based on forward masking, only average data across a number of subjects have been considered. The present study is concerned with bandwidth estimates in simultaneous and forward masking in individual normal-hearing subjects. In order to investigate the reliability of the individual estimates, a statistical resampling method is applied. It is demonstrated that a rather large set of experimental data is required to reliably estimate auditory filter bandwidth, particularly in the case of simultaneous masking. The poor overall reliability of the filter estimates was found to be mainly related to the very short tone duration (i.e., 10 ms) that was chosen. Applying 300-ms long tones in simultaneous masking drastically improved the reliability of the filter estimates. The tone duration in forward masking had to be very short to elicit a sufficient amount of masking. Based on extensive data for three subjects, the difference between forward and simultaneous masking es-timates of auditory filter bandwidth was found to be even larger than previously reported, with a bandwidth decrease by a factor of about 1.8 rather than 1.4. The results of the study can be used to optimize the measures of frequency selectivity which is particularly useful when studying consequences of (individual) hearing impairment.
Proceedings of the 2010 Annual Conference of the Australian Acoustical Society, 2010, p. 3213-3219