Psychoacoustics linked to Plomp’s attenuation and distortion components: Why are simple concepts so complex to predict?
* Presenting author
Abstract:
According to Plomp (1978), speech recognition thresholds (SRTs) in quiet and in noise can simply be described with the Attenuation component A (linked to---and predicted by---the audiogram), and the distortion component D (related to supra-threshold processing deficits). While no clear prediction method of D was available for decades, the FADE model (Framework for Auditory Discrimination Experiments) was used as late as 2018 to successfully infer supra-threshold model parameters predicting D from tone-in-noise detection thresholds. To confirm the feasibility of this approach for a clinical population, we predicted D from tone-in-noise detection thresholds of 93 subjects with normal and impaired hearing, while fits of measured SRTs provided empirical individual A and D-values. Unsurprisingly, multiple linear regressions identify the average audiogram as the best predictor for A. However, no simple relation between D and the psychoacoustic thresholds were found, indicating that a more complex estimation method of D has to be applied, i.e., non-linear regression methods or a model-driven separation of the two components such as FADE. Hence, threshold-based prediction models (like, e.g., SII, STOI or HASPI) cannot support Plomps simple A+D description. This explains why Plomps simple concept was not backed up by appropriate models for decades.