Development of an adaptive scaling method for subjective listening effort

TLDR

The study introduces an adaptive Categorical Listening Effort Scaling procedure that adjusts the signal‑to‑noise ratio to obtain subjective listening‑effort ratings. The method estimates each listener’s SNR range for effort ratings, then presents sentences at random SNRs within that range and fits one or two linear regressions to model effort as a function of SNR. The adaptive procedure produces results independent of the starting SNR, matches static methods while eliminating lengthy pretests and bias, resolves individual and masker differences, and shows high test‑retest reliability (ICC ≈ 0.9) with inter‑individual variability roughly three times intra‑individual variability.

Abstract

An adaptive procedure for controlling the signal-to-noise ratio (SNR) when rating the subjectively perceived listening effort (Adaptive Categorical Listening Effort Scaling) is described. For this, the listening effort is rated on a categorical scale with 14 steps after the presentation of three sentences in a background masker. In a first phase of the procedure, the individual SNR range for ratings from “no effort” to “extreme effort” is estimated. In the following phases, stimuli with randomly selected SNRs within this range are presented. One or two linear regression lines are fitted to the data describing subjective listening effort as a function of SNR. The results of the adaptive procedure are independent of the initial SNR. Although a static procedure using fixed, predefined SNRs produced similar results, the adaptive procedure avoided lengthy pretests for suitable SNRs and limited possible bias in the rating procedures. The adaptive procedure resolves individual differences, as well as differences between maskers. Inter-individual standard deviations are about three times as large as intra-individual standard deviations and the intra-class correlation coefficient for test-retest reliability is, on average, 0.9.