Digital signal processing hearing aids may provide possibilities for new signal processing strategies to compensate for hearing loss. However, to be practically usable in a headworn hearing aid the digital circuitry needs to fulfil requirements of low power consumption, low supply voltage and small size.

This dissertation deals with the design, implementation, utilization and evaluation of a digital hearing aid based on a filterbank. To obtain low power consumption in an ear level digital hearing aid, the filter bank was designed with high order filters for a selective channel separation, but still of low computational complexity, thanks to most filter coefficients (multiplications) being designed with zero-value, not being necessary to execute. The filterbank was utilized in a general signal processing algorithm together with two flexible compressors in a low frequency (LF) and a high-frequency (HF) channel. In order to adjust signal processing to an individual hearing loss configuration, three different fitting algorithms were developed: LinEar, DynEar and RangeEar. All three algorithms provided individual frequency shaping via the filterbank. RangeEar and DynEar used wide dynamic range syllabic compression in the LF channel, while LinEar used compression limiting. In the HF channel, RangeEar used a slow acting automatic volume control, while DynEar and LinEar used compression limiting. The subjects had access to a manual volume control when using the LinEar or DynEar options whereas RangeEar was fully automatic.

Wearable DSP based experimental instruments were used to evaluate the fitting algorithms under real world conditions. Comparative field tests were carried out with experienced hearing aid users. A commercial ear level digital hearing aid based mainly on one of the fitting algorithms (RangeEar) was also evaluated and compared to conventional analog hearing aids. Evaluation included laboratory testing of speech recognition in noise and sound quality ratings. Questionnaires and interviews were used to assess benefit, handicap and disability, and preference of an algorithm. Seven experiments were performed, resulting in a total of about 100 man-months of field experience.

Results with the experimental wearable instruments did not indicate one general good-for-all algorithm, but different algorithms resulting in preference and performance depending on the hearing loss configuration. Sound quality ratings of overall impression, clearness and possibly speech recognition in noise were important factors determining preference for an algorithm. Results indicate that the preference for an algorithm can be predicted from auditory dynamic range data alone. LinEar preference subjects had a relatively flat auditory dynamic range response, whereas DynEar and RangeEar preference subjects had more sloping configurations. It was hypothesized that the different preferences were affected by different susceptibility to masking of high frequency sounds by amplified low frequency sounds.

Results with the ear level digital test aid indicated that the test aid provided significantly more benefit, less experienced handicap, better speech recognition in noise, and clearer sounds as compared to conventional analog aids. Regarding general preference, 80 % of the subjects, stating a certain preference, preferred the test aid as compared to their current well fitted aid.