Talker identification in three types of background noise

Abstract

Perceiving speech under informational masking requires listeners to distinguish and attend to one talker from among competing talkers. However, it is unknown how well listeners are even able to identify a target talker in the presence of noise, much less how this ability is affected by factors such as noise type, signal-to-noise ratio (SNR), and speech content. In a two-day training paradigm, young adults with normal hearing learned to identify five talkers by the sound of their voice in quiet. We then tested talker identification accuracy during masking by speech-shaped noise, multi-talker babble, and a single, unfamiliar competing talker. Test stimuli contained both trained and novel sentences and were presented at three different SNRs. Talker identification was highest in quiet and fell as a function of SNR for all noise types. The change in accuracy as a function of SNR was greatest for multi-talker babble. Across all noise types and SNRs, talkers were identified more accurately from trained sentences than untrained ones. There was substantial individual variability in talker identification accuracy in noise, which appeared unrelated to listeners’ pure-tone hearing levels, speech perception in noise, phonological awareness, or phonological memory, even after controlling for talker identification accuracy in quiet.