A non-intrusive objective speech intelligibility metric tailored for cochlear implant users in complex listening environments

In this work, we propose a speech intelligibility metric tailored for CI users, termed SRMR-CI. This metric is based on the speech-to-reverberation modulation energy (SRMR) measure, which has been shown to correlate well with speech quality and intelligibility for normal hearing listeners. Our proposal extends SRMR to improve its correlation to CI speech intelligibility based on three main modifications. First, we replaced the acoustic filterbank used in SRMR by a CI-inspired filterbank, similar to the one used in the speech coding strategy present in Nucleus CI devices. Adjustment of the filterbank alone led to improvements of 21% in linear correlation, 6% in rank correlation, and 14% in sigmoidal-mapped correlation between the objective and subjective scores when compared to the original SRMR. The second modification was an adjustment to the modulation frequency range used by the metric to compute energies in the modulation spectrum, in order to reduce the metric’s sensitivity to the fundamental frequency of speech. Lastly, we propose a modulation energy thresholding scheme to account for variability related to speech content. These two additional steps can bring additional gains of 19% in correlations and a relative decrease of 50% in the root-mean-square error of the metric predictions. We evaluated the proposed metric and several state-of-the-art metrics using subjective cochlear implant listening ratings for speech in noisy, reverberant, and non-linearly enhanced conditions. The proposed metric showed performance in par with intrusive metrics under environmental distortion conditions and outperformed all the benchmark metrics when assessing the intelligibility of non-linearly processed speech. A non-intrusive metric such as SRMR-CI is a valuable resource for developing and validating new cochlear implant devices and enhancement algorithms, and can potentially be used in devices to adjust system parameters on the go.