Talker variability in real-life speech recognition by cochlear implant users

Abstract

For adult cochlear implant (CI) users, understanding speech in the real world, outside the clinic, can be challenging. A fundamental attribute of real-life speech is the immense amount of talker variability, present from multiple talkers with diverse linguistic and developmental histories. Talker variability, beyond background noise or competing talkers, contributes greatly to real-life, adverse listening conditions. Normal-hearing listeners are able to quickly and efficiently adapt to and learn talker differences to facilitate speech recognition in these conditions. However, talker variability may be particularly detrimental to CI users, since they must rely on signals that are inherently degraded, missing spectro-temporal details. In three experiments, we examined the impact of different sources of talker variability (talkers, accents, speaking styles) on CI speech recognition. The results showed that CI users have difficulty recognizing speech with vast talker variability, and discriminating talkers’ voices and accents, with a great amount of individual differences in performance. These findings suggest that the well-controlled, low-variability conditions in the lab or clinic may not reflect actual CI users’ performance in daily life. [Funding: VIDI Grant (No. 016.093.397) from the Netherlands Organization for Scientific Research (NWO), the Netherlands Organization for Health Research and Development (ZonMw).]