Predicting performance on the digits-in-noise test in noise and reverberation using the speech transmission index.

Abstract

This study investigated the effects of noise and reverberation on recognition of Dutch digits-in-noise (DIN) test triplets. The findings demonstrate that the STIDIN, an adaptation to the standard speech transmission index (STI), provides a reliable, low‑error predictor of DIN test performance in conditions with noise and/or reverberation when only a single speech recognition threshold (SRT) measurement in noise is completed. To achieve this, twenty-four normal-hearing (NH) adults completed adaptive SRT measurements in noise-only, reverberation-only and combined noise and reverberation conditions using unprocessed (UP), low-pass filtered (LPF) and cochlear implant vocoded (CIvoc) speech materials. Standard STI values differed across conditions and overestimated the detrimental effect of reverberation on recognition. The STIDIN, derived from the magnitude cross power spectrum (mCPS) of reverberated DIN test triplets and unprocessed triplets, was applicable for T60 reverberation times up to 12 s. STIDIN values remained constant across conditions, showing no significant main effect of condition for any of the unprocessed and processed speech materials. The results show that when an SRTn measurement in noise-only is obtained using DIN test triplets (i.e., the clinical standard), the STIDIN can be used to predict the SRT in other listening conditions with noise and reverberation. The root‑mean‑square error between measured and predicted SRTs was 0.92 dB, and 95 % of predictions deviated <1.7 dB from the measured values in conditions with noise and/or reverberation.