Thomas Lunner

Research Area Manager / Adj. Professor

> Learn more

Evaluating the concept of User Preference Fitting

We evaluated the concept of User Preference Fitting (UPF), where participants adjusted the hearing aid settings themselves. Subgroups indicated that UPF fittings can be very positive, resulting in improved performance, or very negative, resulting in poorer performance compared to prescribed settings.

Testing User Preference Fitting on experienced hearing-aid users

One way to individualise hearing aid fittings would be to give users a ‘steering-wheel’ or fine-tuning trimmers, enabling them to adjust the fitting parameters to their own preference in given listening environments (UPF). This was facilitated by means of a neural network. This network was trained to individual user preference settings in 11 different pre-defined environments – such as a post office or cafeteria - where two people were conversing in realistic background sounds.

18 experienced hearing-aid users participated in this study. In a field test, subjects were given access to both prescribed settings and self-adjusted settings in a triple-blind test design. Outcome measures were speech reception threshold (Hagerman sentences) in unmodulated and modulated noise; success or failure with the user preference fittings; subjective preferences (blinded settings), and themes from a semi-structured interview.


On average there were no differences in speech reception thresholds (SRTs) between UPF and Prescribed Settings. The blinded preferences of UPF versus Prescribed Settings did not show a pattern of preference for either of the two settings. But as a whole, the individual subjective preferences, SRTs, and interview outcomes revealed an interesting pattern of results.

One subgroup of subjects (about one third) demonstrated consistently better performance with the UPF setting compared to the Prescribed settings. This supports the idea that hearing-aid settings are important for this subgroup. Compared to the others, these groups found it easier to listen and accept the consequences of making adjustments to improve the fitting. However, what makes these subjects special is unknown. It could be that their absolute listening abilities were better – like skilled musicians - but this was not investigated in this study. 

Another subgroup of the subjects (about one third) failed in their UPF setting, resulting in unrealistic neural network patterns. This led to poor performance and rejection of the self-adjustment fittings. They also expressed uncertainty about the adjustments - relying more on the prescribed settings – and appeared to be less confident about the decisions they were making.

Evaluating the concept of user preference fitting 1 Consequence of UDF fittings

The consequence of UPF fittings

The results among the subgroups indicate that the consequence of UPF fitting is reinforcement. Either the fitting can be very positive, resulting in improved outcomes - or very negative, resulting in poorer performance compared to prescribed settings. Therefore, to be able to use self-adjustments in a clinical setting, the direction of reinforcement for the individual client must be predicted, with self-adjustments only being provided to the successful subgroup.

Evaluating the concept of user preference fitting 2 Experimental UPF hearing aid

Experimental UPF hearing aid

The basic DSP principles of the UPF hearing-aid have inspired recent Oticon hearing-aid platforms. The basic principle of the experimental UPF hearing-aid had three functional blocks: Filtering, Analysis, and Control:

  1. Analysis: We used signal level and spectral slope implemented by measures of signal energy for low and high frequencies (the split frequency is 1500 Hz).
  2. Control: We used a non-linear multi-dimensional function in terms of a neural network. The non-linear mapping is a multi-dimensional filter characteristic as a function of the input from the Analysis block i.e. the low and high frequency energy of the input signal.
  3. Filtering: We used a symmetrical FIR-filter that was controlled by the non-linear mapping.

11 different listening environments were used, with background levels ranging from a library (soft) to a bottling hall (very loud). In all listening environments, two people were conversing in the given background to get vocal effort correct, so the listener could ‘calibrate’ his/her expectations of the environment. The trained network then generalised the preferred settings under real-world listening environments.


Further reading

Dreschler WA, Keidser G, Convery E, Dillon H (2008) Client-based adjustments of hearing aid gain: The effect of different control configurations. Ear and Hearing, 29(2), p. 214-227.

Zakis JA, Dillon H, McDermott HJ (2007) The design and evaluation of a hearing aid with trainable amplification parameters. Ear and Hearing, 28(6), p. 812-830.

Elberling C, Hansen KV (1999) Hearing instruments - interaction with user preference. Proceedings of the 18th Danavox Symposium, p. 341-357.

Learn More

  • learnmore-poster


    Lessons from the use of self-adjustment in a research setting – User Preference Fitting