Data-driven auditory profiling as a tool to define Better hEAring Rehabilitation (BEAR)

  • Raul Sanchez Lopez (Guest lecturer)
  • Michal Fereczkowski (Guest lecturer)
  • Federica Bianchi (Guest lecturer)
  • Santurette, S. (Guest lecturer)
  • Dau, T. (Guest lecturer)

    Activity: Talks and presentationsConference presentations


    While the audiogram still stands as the main tool to select hearing-aid compensation strategies in audiological clinics, there is ample evidence that loss of hearing sensitivity cannot fully account for common difficulties encountered by people with sensorineural hearing loss, such as understanding speech in noisy environments. Forty years after R. Plomp proposed his attenuation-distortion model of hearing impairment, it remains a challenge to address the distortion component, mainly related to supra-threshold deficits, via adequate clinical diagnostics and corresponding hearing-aid compensation strategies. Inspired by the different auditory profiling approaches used in the literature, one aim of the Better hEAring Rehabilitation (BEAR) project is to define a new clinical profiling tool, a test-battery, for individualized hearing loss characterization.
    The proposed BEAR approach is based on the hypothesis that any listener’s hearing can be characterized along two dimensions reflecting largely independent types of perceptual distortions. In order to keep the approach as neutral as possible, no a priori assumption was made about the nature of the two distortion types. Instead, a statistical analysis method, combining unsupervised and supervised learning, was developed to learn from existing data. The aim was to provide a tool to help define the two distortion types, such that potentially relevant tests to classify listeners into different auditory profiles could be identified. So far, the data from previous auditory profiling studies were reanalyzed based on this approach. First, an unsupervised-learning technique including archetypal analysis was used to identify extreme patterns in the data, forming the basis for different auditory profiles. Next, a decision tree was determined to classify the listeners into one of the profiles.
    The data-driven analysis provided consistent evidence for the existence of two independent sources of distortion, and thus different auditory profiles, in the data. The results suggested that the first distortion type was related to loss of sensitivity at high frequencies as well as reduced peripheral compression and frequency selectivity, while the second distortion type was linked to binaural temporal-fine-structure processing abilities as well as low-frequency sensitivity loss. The audiogram was not found to reflect an independent dimension on its own, and the most informative predictors for profile identification beyond the audiogram were related to temporal processing, binaural processing, compressive peripheral nonlinearity, and speech-in-noise perception. The current approach can be used to analyze other existing data sets and may help define an optimal test battery to achieve efficient clinical auditory profiling.
    Period16 Aug 2018
    Event titleInternational Hearing Aid Conference 2018
    Event typeConference
    LocationTahoe, United States, CaliforniaShow on map