Advanced Computational Models of Hearing Loss
Models of hearing loss are often used in the clinic or as a research tool to stand in for a real patient when fitting hearing devices or developing hearing-aid signal processing algorithms. To the extent that such tools are accurate, they can save a great deal of testing time, because optimal settings can be arrived at without testing and re-testing the patient’s hearing abilities. Most models of hearing loss and speech intelligibility make use of basic clinical data – the patient’s audiogram – to arrive at a speech intelligibility prediction. Such models have been incorporated into everyday clinical use to optimize hearing-aid parameters to maximize speech intelligibility for a given patient. The models are nevertheless only partly successful, and are not able to predict variability in performance across individual patients with similar clinical audiograms.
A unique feature of our research in this area is the focus on individual differences in speech-recognition capabilities. Each individual subject is characterized along a number of different dimensions, from basic auditory and visual acuity for the simplest speech elements, to the cognitive processes engaged in interpreting complex sentence structures. In this way, a profile is established for each individual patient, providing a basis for rehabilitation. Based on these profiles, computational models are constructed to predict speech recognition more accurately than models based on the clinical audiogram alone.