Effect of noise and reverberation on speech intelligibility with cochlear implants considering realistic sound environments
thesisposted on 28.03.2022, 09:16 by Javier Badajoz-Davila
Understanding speech in background noise and reverberation is a known problem for individuals with cochlear implants (CIs). However, current laboratory-based assessments of speech intelligibility (SI) are usually poor predictors of an individual's listening abilities in the real-world. This mismatch is largely attributed to the use of oversimplified methods whereby neither the speech nor the noise employed in the tests resemble what an individual experiences in their daily life. In order to better understand the challenges faced by individuals listening with CIs, the present study employs materials and methods designed to bridge the gap between laboratory based outcomes and individuals' experience of everyday listening. This work comprises three main studies. The first two studies aim to systematically evaluate and understand the effect of realistic reverberation on SI in CI recipients. Sentence recall performance was measured in 12 unilateral CI recipients in both quiet and noise, considering six realistic rooms at varying target-to-receiver distances. The results suggest that in quiet conditions reverberation has a significant impact on SI mainly at long distances, with the exception of small reflective rooms, where SI is affected even at close, conversational distances. Further analysis of the data in quiet conditions suggests that room acoustic parameters such as the U50 can predict SI in rooms with reasonable accuracy. Analysis in noise revealed that the temporal smearing effect of reverberation on the noise signal is beneficial to SI, an effect that is not accounted for by the U50. Hence, future implementations of the U50 need to consider the noise-inherent modulations. The goal of the third study is twofold. First, to understand the effect of test realism on SI outcomes. Second, to assess SI performance as well as bilateral benefit in CI recipients in more realistic noisy conditions. Sentence recall performance was measured in 15 bilateral CI recipients using sentence materials as well as noises with different level of realism. "Standard" BKB-like sentences were used as well as more realistic sentences that were cut out of natural two-talker conversations elicited at different vocal effort levels. Both sentence materials were presented in different realistic acoustic environments at natural signal-to-noise ratios (SNRs) as well as in "standard" babble noise. The results indicate that participants could more easily deal with babble noise than with more realistic noisy situations, and that they could understand more easily the standard sentences than more realistic (conversational) speech. This effect was pronounced at lower SNRs. A small but significant bilateral benefit was observed in most conditions. The present work highlights the importance of using realistic reverberation, presentation levels, speech material, noise material and spatial representation of the sound field when assessing SI performance in CI recipients -- abstract.