Optimalisation of hearing aids through paired comparisons
Researcher(s): B.A.M. Franck, PhD.
Supervisor(s): prof. W.A. Dreschler, PhD.
Background:
Fitting strategies in current clinical practice are often sequential. This means first one parameter is estimated and then the next parameter. Implicitly it is assumed that there are no or very few interactions between the different parameters. The justification of this assumption is questionable. To gain more insight into the interactions between hearing aid parameters research on interactive hearing aid fitting is needed. In interactive fitting several hearing aid parameters are adjusted simultaneously including interactions between the parameters.
Aim:
The aim of this project is to develop an interactive hearing aid fitting strategy that leads to an individual fit of the complex hearing aid algorithm. The study focused on two issues. What are the merits of an adaptive search strategy (Simplex)? And what are the results for different conditions? The first issue comprises methodological issues as reliability (test – retest), consequences of variation of the starting point, influence of the evaluation criterion and the interaction between the algorithms. The second issue focused on the merits of individual fitting from an audiological point of view. Does the optimal setting depend on the individual? Does it depend on the background noise?
Method:
The fitting procedure is based on paired comparisons of sentences. Optimalisation based on paired comparisons can efficiently be performed by the use of the Simplex method. This method is determines the largest gradient in order to approach the optimum as fast as possible. It turns out to be suited to differentiate between different settings in multiple dimensions. Because the hearing impaired subject can participate in the hearing aid fitting the method has a positive psychological effect (Kuk, 1994). In our studies we adapted the Simplex method in several ways. First we applied the Simplex method for three dimensions. And in contrast to previous studies we didn’t investigate three parameters of one algorithm, but we investigated three different algorithms.
Results:
From our studies several conclusions can be drawn. The test-retest reliability is largest for listening comfort. The arbitrariness that emerges in the judgement of speech intelligibility stems from (too) little perceptual differences and the preference for the second fragment. The consistency has shown to be low because of unfavourable response distributions. Differences in optimal settings between different subjects, different evaluation criteria and different conditions argue in favour of an adaptive search strategy. The outcome suggest that the Simplex method must only be applied if several requirements are met:
- variable step size in the Simplex procedure for each direction;
- perceptive different settings per algorithm;
- no algorithms with response distributions that are flat or contain multiple optima;
- a possibility to repeat fragments and maybe to indicate ‘no difference’;
- presentation of sentences as test material.
Only if these requirements are met, the test-retest reliability and the consistency of the optimal setting for different starting points may be sufficient. The advantage of the method is that interaction between signal processing algorithms are taken into account. Besides with the method of paired comparisons listening comfort can be subjectively measured. Together with the fact that the listener becomes more involved in the fitting process this improves the acceptation of the final setting.
Future research:
This research has led to the thesis of dr. B.A.M. Franck, titled ‘Hearing-aid fitting in interaction – On optimal combinations of multiple acoustic signal-processing strategies’. He received his PhD on June 29th 2004. Research on this topic will be continued as part of the HEARCOM-project.
