Effects of talker variability in cochlear implants

Speech perception is influenced by many factors; differences in talker characteristics also significantly influence speech perception. Because speech understanding is fairly robust to variations in pronunciation, researchers have theorized that a "speaker normalization" process occurs, in which multiple talkers' speech patterns are normalized to a target pattern. However, effects of "talker variability" may also be produced, in which speech recognition performance worsens with multiple talkers, relative to that with a single talker. Due to partial loss of talker voice information in spectrally degraded speech as experienced by cochlear implant users, the talker variability effects in cochlear implant speech may be more complex and the underlying mechanism is still not clear.;The first part of this study is perceptual experiments investigating talker variability effects in cochlear implants. A baseline experiment with cochlear implant users showed a significant talker variability effect. Then a series experiments was conducted with normal hearing subjects in simulations to investigate effects of number of talkers, talker set, background noise, carrier, spectral resolution, and number of successive talker-specific stimuli.;In the second part of this study, we compared talker variability effects with talker discrimination performance. The results revealed a strong correlation between talker variability effects and talker discrimination performance. The most prominent talker variability effects were observed in cochlear implant users with moderate levels of talker discrimination performance. Talker variability effects were relatively small in cochlear implant users with relatively good or poor talker discrimination performance.;The third part of this study is acoustic analyses and modeling of talker variability effects in cochlear implants. The acoustic distance between vowel stimuli and scatter indices of vowel pairs were calculated based on the temporal spectral envelope of the vocoder processed vowel stimuli. To account for talker variability effects, modifications were made respectively for single- and multi-talker context. In general, the normalized average acoustic distance could well predict the overall performance and talker variability effects. However, the prediction for the detailed confusion patterns showed some mixed results. The results are consistent with the assumption that talker variability effects are caused by listeners' adaptation to talker characteristics in a single-talker context.