The role of talker acoustics for intelligibility and effort in adverse listening conditions
A current focus in hearing research is the use of measures of listening effort to complement speech audiometry (Pichora-Fuller et al., 2016). Central to research into listening effort is the emphasis on the characteristics of the listener such as nativeness or hearing status. In this study, we investigated the effect of talker acoustics on listening effort by measuring the relative importance of various acoustic-phonetic dimensions in adverse listening conditions. Previous research related talker-specific acoustic-phonetic features to intelligibility in noise (Hazan & Markham, 2004), but has not yet accounted for listening effort.
Based on an anechoically recorded corpus of sixteen Southern British English speakers, we conducted listening experiments in combination with pupillometry. We presented temporally or spectrally distorted speech (using time-compression and noise-vocoding, respectively). Furthermore, undistorted speech was presented in clear and with speech-shaped masking noise. Intelligibility scores were obtained based on keywords recognised correctly. Listening effort was measured by tracking pupil size changes over time.
Our results are in line with previous studies measuring talker intelligibility in noise. Pupil dilation was increased for both, distorted and masked speech. We found that speaking rate was a common predictor of intelligibility in both distorted conditions. We furthermore observed that speaking rate was related to changes in pupil size, indicating reduced listening effort and sustained retrieval processes for slower talkers. The current results are interpreted in the context of models of listening effort such as FUEL (Pichora-Fuller et al., 2016).