A binaural model predicting speech intelligibility in noise for hearing-impaired listeners

A binaural model is presented which predicts the effect of audibility on the intelligibility of speech in the presence of speech-shaped noise and vocoded speech maskers. It is based on the short-term binaural speech intelligibility model described by Collin & Lavandier [J. Acoust. Soc. Am. 134, 1146-1159 (2013)] and takes the calibrated target and masker signals (independently) at each ear as inputs along with the listener hearing thresholds in order to calculate a binaural “effective” signal-to-noise ratio. Differences in ratio across conditions can be directly compared to differences in speech reception threshold (SRT).
Model predictions are compared to SRTs measured in the presence of two speech-spectrum noises or two vocoded-speech maskers, which were either (artificially) spatially separated or co-located with the frontal speech target. The spatial separation was realized by presenting each masker on a different single ear using headphones, while the target was presented diotically as coming from the front. Comparing the co-located and separated configurations allows evaluating a spatial release from masking (SRM) which was based here primarily on better-ear glimpsing (no realistic ITDs simulated). Normal-hearing and hearing-impaired listeners were involved in the data collection, which mostly used stimuli spectrally shaped for equalized audibility across listeners. Audibility was varied during the experiment by testing four broadband sound levels for the combined maskers (while the target level was varied relative to these reference levels to measure the SRTs).
For both group of listeners, the model allows a good prediction of the decrease of SRT as well as the increase of SRM with increasing audibility/levels. The averaged absolute error between data and prediction was generally below 1 dB across tested conditions.