Auditory model-based parameter estimation and selection of the most informative experimental conditions

• Published in: Acta acustica Vol. 8; p. 3
• Main Authors: Dietze, Anna, Reinsch, Anna-Lena, Encke, Jörg, Dietz, Mathias
• Format: Journal Article
• Language: English
• Published: Stuttgart EDP Sciences 2024
• Subjects: Accuracy; audiological diagnostics; binaural hearing; computational audiology; Data collection; Experiments; Mathematical models; Measurement techniques; Model accuracy; model-based experiment steering; Parameter estimation; Parameter identification; Pathology; Performance prediction; Physiology; Steering; Time measurement; tone-in-noise detection
• ISSN: 2681-4617; 1610-1928; 2681-4617; 1861-9959
• DOI: 10.1051/aacus/2023064

Identifying the causes underlying a person’s hearing impairment is challenging. It requires linking the results of listening tests to possible pathologies of the highly non-linear auditory system. This process is further aggravated by restrictions in measurement time, especially in clinical settings. A central but difficult goal is thus, to maximize the diagnostic information that is collectable within a given time frame. This study demonstrates the practical applicability of the model-based experiment-steering procedure introduced in Herrmann and Dietz (2021, Acta Acustica, 5:51). The approach chooses the stimuli that are presented and estimates the model parameters best predicting the subject’s performance using a maximum-likelihood method. The same binaural tone-in-noise detection task was conducted using two measurement procedures: A standard adaptive staircase procedure and the model-based selection procedure based on an existing model. The model-steered procedure reached the same accuracy of model parameter estimation in on average only 42% of the time that was required with the standard adaptive procedure. Difficulties regarding the choice of a reliable model and reasonable discretization steps of its parameters are discussed. Although the physiological causes of an individual’s results cannot directly be inferred using this procedure, a characterization in terms of functional parameters is possible.