Most existing speech intelligibility measures are either designedfor single-channel applications – hence unsuited to evaluatehearing aid algorithms –or intrusive, applicable only in simulated scenarios in which the clean signal is available. Non-intrusive speech intelligibility measures able to reliably predictspeech intelligibility without knowledge of the clean signal areurgently needed. This paper proposes a non-intrusive measurethat predicts speech intelligibility using only the processed signals and audiogram of the listener as input. The proposed measure relies on three steps, namely a hearing-loss model, a feature extractor and a predicting function. The hearing loss modeluses the target signal and the listener’s audiogram as input whilethe feature extractor and the predicting function are trained onprocessed signals labeled in terms of speech intelligibility during a listening test. The evaluation is conducted using cross-validation on both tracks of the first Clarity Prediction Challenge (CPC1).