Objective Prediction of Hearing Aid Benefit Across Listener Groups Using Machine Learning: Speech Recognition Performance With Binaural Noise-Reduction Algorithms.

The model was tested against measurements from one previous study, not against a new clinical sample. It was evaluated mainly with the German matrix sentence test, not with other speech tests. The single interfering talker condition was not suitable for prediction. Individual predictions for impaired-hearing listeners were based only on audiograms and did not capture all individual factors. The predictions generally underestimated measured performance, so absolute thresholds were biased. The comparison used 10 normal-hearing listeners and 12 aided impaired-hearing listeners, so it was not a large clinical trial.

Do not use this model to choose a hearing aid noise-reduction setting for an individual patient. It predicted group benefit, but individual benefit predictions based only on audiograms were weak.