A series of SNR-based speech intelligibility models in the Auditory Modeling Toolbox

Abstract

This technical paper presents a series of speech intelligibility models that have been developed since the original version proposed by Lavandier and Culling [(2010). Journal of the Acoustical Society of America 127, 387–399]. This binaural model accounts for better-ear listening and binaural unmasking to predict the intelligibility of a near-field target speech among multiple stationary noise sources in rooms for normal-hearing listeners. Subsequent model versions allowed to consider a reverberated speech target in the far-field, envelope-modulated noise sources, and hearing-impaired listeners. As an intermediate step before considering speech maskers, a monaural version incorporating a harmonic-cancellation mechanism was recently developed to account for the effect of a stationary harmonic masker. This technical review is oriented towards model users and explains when and how each model should be used, points at its advantages and limitations, and provides an example of predictions using a data set from the literature. All these models along with the data, signals and code used to prepare the presented figures are made available within the Auditory Modeling Toolbox (AMT 1.1).