Pupillometry Reveals the Role of Signal-to-Noise Ratio in Adaption to Linguistic Interference Over Time

Studies of speech-in-speech listening show that intelligible maskers are more detrimental to target perception than unintelligible maskers, an effect we refer to as linguistic interference. Research also shows that performance improves over time through adaptation. The extent to which the speed of adaptation differs for intelligible and unintelligible maskers and whether this pattern is reflected in changes in listening effort are open questions.

In this preregistered study, native English listeners transcribed English sentences against an intelligible masker (time-forward English talkers) versus an unintelligible masker (time-reversed English talkers). Over 50 trials, transcription accuracy and task-evoked pupil response (TEPR) were recorded, along with self-reported effort and fatigue ratings. In Experiment 1, we used an adaptive procedure to ensure a starting performance of ~50% correct in both conditions. In Experiment 2, we used a fixed signal-to-noise ratio (SNR = -1.5 dB) for both conditions.

Both experiments showed performance patterns consistent with linguistic interference. The speed of adaptation depended on the SNR. When the SNR was higher for the intelligible masker condition as a result of the 50% starting performance across conditions (Experiment 1), adaptation was faster for that condition; TEPRs were not affected by trial number or condition. When the SNR was fixed (Experiment 2), adaptation was similar in both conditions, but TEPRs decreased faster in the unintelligible than intelligible masker condition. Self-reported ratings of effort and fatigue were not affected by masker conditions in either experiment.

Learning to segregate target speech from maskers depends on both the intelligibility of the maskers and the SNR. We discuss ways in which auditory stream formation is automatic or requires cognitive resources.