gms | German Medical Science

Within the last years an increasing number of model-based and machine learning approaches for supporting audiological diagnostics and treatments can be noticed. They cover a broad range of applications using for example statistical models for profiling patients and profile-based rehabilitation, perception models that bridge psychoacoustics and speech recognition for individualized characterization of hearing impairment and hearing loss compensation strategies or decision-support systems relaying on big data and machine learning algorithms.

This contribution reports current developments of auditory models focusing on better understanding of the consequences of hearing loss on listeners performance considering different listening conditions and tasks. This includes simulations of automatic speech recognition-based model called the Framework for Auditory Discrimination Experiments (FADE). FADE was developed to predict outcomes of speech recognition tasks and of psychoacoustic experiments. Thresholds collected in empirical studies with the psychoacoustic measurements may be used in the model to individualize different components hearing loss. Due to its constrains, FADE is well suited to simulate speech recognition of hearing-impaired listeners without and with hearing devices.

This contribution focuses on aspects relevant for the application of FADE as a supporting tool in audiology. A few studies will be reported aiming at:

1.

contribution of hearing threshold and supra-threshold deficits to simulated speech recognition in noise in standard laboratory conditions,

2.

application of FADE for assessment of benefit from hearing devices,

3.

model-based selection of experimental conditions in acoustically complex scenes for diagnostical and rehabilitation purposes

Predicted speech recognition is compared to empirical data and by that the benefit from using model-based approaches in audiology is discussed as well as limitations of the current approaches.