Reliability of categorical loudness scaling and its relation to threshold

Abstract

Objective: To further examine the reliability of categorical loudness scaling (CLS) for individual loudness categories and for the slope of the CLS functions. And, to evaluate the relationship between CLS and audiometric threshold.

Design: CLS functions were obtained in 74 subjects, 58 with hearing loss and 16 with normal hearing. CLS functions were measured at three frequencies (1, 2, and 4 kHz) in two separate sessions separated by as little as 1 wk and as much as 6 mo. Reliability of mean and median levels within each loudness category was assessed using SDs and correlation coefficients. Lines were fit to the CLS functions, and slopes of the lines were used to assess reliability and the relation between CLS and audiometric threshold.

Results: Similar reliability for CLS measurements was observed in both normal-hearing and hearing-impaired subjects at all frequencies. Across both groups of subjects, correlations describing the reliability of mean stimulus level within category exceeded 0.92 at all frequencies. In addition, SDs of the mean stimulus-level difference between visits ranged from 6.6 to 7.8 dB, depending on frequency. The correlation between the slope of a straight line fitted to the entire CLS function and audiometric threshold collapsed across frequencies was 0.72. Two line segments were then fit to the CLS function: one segment was fit to the soft portion of the CLS function (categorical units < or =20) and the other segment to the loud portion (categorical unit >20). Slopes of the line fit to the entire CLS function and of the line fit to the soft portion of the CLS function were both reliable across sessions. The slope of the line fit to the soft portion increased as audiometric threshold increased, with the correlations greater than 0.86 at all frequencies. No relationship was observed between slope of the line fit to the loud portion of the CLS function and audiometric threshold. Iso-loudness contours were constructed from the CLS data and used to determine the gain that would be needed to produce "normal" loudness percepts for hearing-impaired individuals.

Conclusions: Within-subject CLS measurements were reliable across sessions both for individual loudness categories and for slope of the CLS functions. In addition, the slope of the low-level portion of the CLS function varied in a predictable manner with audiometric threshold, with slope increasing as audiometric threshold increased. Finally, gain as a function of input level needed to provide loudness percepts for individuals with hearing loss equal to the loudness percepts of normal-hearing individuals can be estimated from audiometric threshold. This finding supports the assumption that audiometric threshold and response growth (loudness) are both determined by the same underlying cochlear mechanisms.

Figure 1

CLS scale with 11 response categories displayed on a touch-screen monitor for use by the subjects to rate the loudness of the signal. The numbers along the left side indicate categorical units (CU) and were used for recording the subject’s responses and data analysis.

Figures 2a and 2b

Figure 2a (top) plots the mean stimulus level from the three trials in session 2 as a function of the mean stimulus level for the three trials from session 1. Each data point represents levels for every individual category used by each subject during CLS measurements. Figure 2b (bottom) plots the slope (CU/dB) of a simple regression line fit to all of the data from each subject’s entire CLS function from session 2 as a function of the slope of the function fit to the comparable set of individual data from session 1. In both figures, data for each frequency are provided separately in each panel and correlations are provided as insets in each panel. Data from normal-hearing and hearing-impaired subjects are combined within each panel.

Figure 3

Slope (CU/dB) of a straight line fit to the entire CLS function for each subject is plotted as a function of audiometric threshold, following the convention used by Launer (1995). Data for different frequencies are represented by different symbols: squares – 1 kHz; circles – 2 kHz; triangles – 4 kHz. The correlation provided within the figure was determined by collapsing across frequency.

Figure 4

Slope (CU/dB) of a line fit to decreasing ranges of categories from averaged individual loudness functions are shown as a function of audiometric threshold. The slopes associated with each range of categories are plotted in separate panels, going from the most inclusive to the least inclusive range (upper left panel and lower right panel, respectively). Within each panel, data are shown for all three frequencies. Correlations are provided as insets within each panel.

Figure 5

Slopes (CU/dB) of the lines fit to the soft and loud portions of each subject’s CLS function as a function of audiometric threshold, with circles representing soft slope and triangles representing loud slope. Data are plotted separately for each frequency.

Figure 6

Iso-loudness contours constructed by finding the median level in dB SPL assigned to each loudness category as a function of audiometric-threshold group with CU as the parameter (shown as different symbols). The lines represent best fits to the data for each CU.

Figure 7

Hypothetical CLS functions for one normal-hearing (circles) and one hearing-impaired subject (squares). The horizontal arrows represent the increase in SPL needed to normalize the loudness percept for the hearing-impaired individual. For ease of visualization, stimulus levels for only half the loudness categories are shown.

Figure 8

The estimate of gain needed to normalize the loudness percept as a function of input level with hearing-threshold categories as the parameter. The gain estimates were obtained from the lines fit to the data in Fig. 6. Data for each frequency are shown in separate panels.