Listeners face various challenges when listening to speech in a background of competing talkers. Behavioural studies of adaptation during speech-in-speech listening show that performance can improve over the course of a test block, with faster improvement in unintelligible compared to intelligible maskers. However, it is unclear whether this pattern is reflected in changes in listening effort.

In this study, native British English listeners undertook a 50-trial speech recognition task in two masker conditions: intelligible maskers (English talkers) and unintelligible maskers (time-reversed English talkers). In Experiment 1 (N=40), participants first undertook an adaptive procedure to find the 50% SRT for each condition. Participants thus started the main task at ~50% correct in both conditions. In Experiment 2 (N=40), the same SNR was used for both conditions (-1.5 dB). In Experiment 3 (planned N=40), participants undertook an adaptive procedure such that they started the unintelligible masker condition at ~41% correct and the intelligible masker condition at ~66% correct. These values represent the mean performance levels from Experiment 2, but switched such that the intelligible masker was presented at the easier SNR and vice versa.

In Experiment 1, performance started around 50% in both conditions as planned. Improvement was faster in the intelligible than unintelligible condition, contrary to our expectations. Pupil dilation decreased over time at the same rate for both conditions. In Experiment 2, performance was higher in the unintelligible than the intelligible condition. Unlike Experiment 1, improvement was faster in the unintelligible than the intelligible condition. Pupil dilation mirrored that pattern, with a faster decrease in the unintelligible than intelligible condition. The data of Experiment 3 are being analysed.

These results suggest that the ease with which listeners learn to stream a target from a competing talker depends on both the intelligibility of the masker and relative SNRs. In Experiment 2, faster transcription improvement and a more pronounced decrease in effort for unintelligible than intelligible maskers suggest that the linguistic interference created by an intelligible masker leads to persistent cognitive demands associated with ignoring the masker’s linguistic content. However, as indicated by Experiment 1, this pattern might arise from an initial performance advantage in the unintelligible masker condition (due to a more favourable SNR) at the start of the task. Results from Experiment 3, in which the SNRs are adjusted to ensure an initial performance advantage in the intelligible masker condition, will help to disambiguate these two interpretations.