Towards predicting binaural audio quality in listeners with normal and impaired hearing
About
Eurich et al. (2024) recently introduced the computationally efficient monaural and binaural audio quality model (eMoBi-Q). This model integrates both monaural and binaural auditory features and has been validated across six audio datasets encompassing quality ratings for music and speech, processed via algorithms commonly employed in modern hearing devices (e.g., acoustic transparency, feedback cancellation, and binaural beamforming) or presented via loudspeakers. In the current study, we expand eMoBi-Q to account for perceptual effects of sensorineural hearing loss (HL) on audio quality. For this, the model was extended by a nonlinear auditory filterbank. Given that altered loudness perception is a prevalent issue among listeners with hearing impairment, our goal is to incorporate loudness as a sub-dimension for predicting audio quality in both normal-hearing and hearing-impaired populations. While predicting loudness itself is important in the context of loudness-based hearing aid fitting, loudness as audio quality sub-measure may be helpful for the selection of reliable auditory features in hearing impaired listeners. The parameters of the filterbank and subsequent processing stages were informed by the physiologically-based (binaural) loudness model proposed by Pieper et al. (2018). This study presents and discusses the initial implementation of the extended binaural quality model.