Fig. 3

Scheme illustrating self-supervised prediction of hearing thresholds from evoked ABR signals. A Stacked ABR time series for different given sound pressure levels (SPL) at a given stimulation frequency, e.g. 18 kHz. The arrow indicates the manual hearing threshold HT\(_{\hbox {G}}\), which is only used for validation in this approach. B Amplitude signal in the time domain for a single SPL. C Amplitude in the frequency domain after Fourier transformation (FT). The lowest 50 frequencies are extracted from the power spectrum as features. D estimated vs. measure SPL. Blue dots: sound pressure levels estimated by random forest regression plotted against the actual measure sound levels. Orange line: a piece-wise function composed of a constant (I) and a polynomial (II) part. The polynomial part of the function is fitted using an elastic net. The circle indicates where the constant and polynomial function meet, which determines the predicted hearing threshold HT\(_{\hbox {P}}\)