Individual differences in simultaneous masking with random-frequency, multicomponent maskers

Individual differences were examined for 49 listeners in a simultaneous-masking task with random-frequency, multicomponent maskers and a 1000-Hz signal. Across conditions, the maskers were broadband noise or complexes of 2-100 sinusoids whose frequencies were randomly chosen with each presentation. Maskers were equal rms waveforms, presented at 60 dB SPL. Thresholds for the multicomponent maskers spanned a range of up to 59 dB across listeners within conditions without distinct groupings. When divided into quartiles, mean thresholds for listeners in the lowest quartile increased monotonically with the number of masker components and peaked below values produced by broadband noise. Listeners in the upper quartile had nonmonotonic functions with a broad peak around 10-20 components, with values above those produced by broadband noise. Quiet thresholds and masked thresholds for the broadband-noise masker did not distinguish "high-" from "low-threshold" groups. Lowering masker uncertainty (frequencies randomized between but not within trials) or introducing temporal differences by shortening the signal both improved performance by as much as 40 dB, dependent on the masking produced with maximum uncertainty. Training effects within experiment (excluding an initial training period) were examined for all listeners and for five listeners across two consecutive experiments. Eleven of forty-nine listeners (22%) showed systematic improvement averaging 12 dB within experiment, independent of high or low original thresholds. For the five tested longer, four improved by 5 dB or more in at least one condition, but the change in performance occurred primarily during initial training for the second experiment. The large individual differences documented present a challenge to the development of models for masking produced by masker-frequency uncertainty.