Faculty Bibliography

It is hypothesized that for postlingually deafened adult cochlear implant (CI) users, a significant source of their perceptual performance variability is attributable to differences in their ability to discriminate the basic perceptual cues that are important in speech recognition. Previous research on 'electric hearing' has identified consistent perceptual cues for vowel recognition. However, the results on consonant perception by CI users are less clear. The primary purpose of this study is to present a quantitative method of evaluating potential 'electric cues' used by CI users in consonant identification. Since the actual input signals to the auditory periphery of CI users are electric in nature, we elected to measure the CI electric discharge patterns in addition to the original acoustic waveforms. The characteristics of the electric discharge patterns in response to intervocalic consonants were quantified and correlated with the dimensions of CI patients' perceptual spaces, which were computed from multidimensional scaling analyses of their consonant confusion matrices. The results agree with most, but not all, commonly accepted acoustic cues used by normal-hearing listeners. The correlation findings also suggest that CI users employ different sets of 'electric cues' in perceiving consonants that differ in their manner of articulation. Specifically, spectral and temporal cues associated with slowly changing formant structures and transitions, and features associated with frication and high-frequency noise, are all highly correlated with the perceptual dimensions of all CI users. However, rapidly changing formant transitions, such as those present in stop consonants, did not appear to play a significant role in consonant recognition by more poorly performing CI subjects. The perceptual results were consistent with our physical findings that the SPEAK coding strategy partially degraded the rapidly changing formant transitions

The purpose of this study was to determine whether 2 speech measures (peak intraoral air pressure [IOP] and IOP duration) obtained during the production of intervocalic stops would be altered as a function of the presence or absence of auditory stimulation provided by a cochlear implant (CI). Five pediatric CI users were required to produce repetitions of the words puppy and baby with their CIs turned on. The CIs were then turned off for 1 hr, at which time the speech sample was repeated with the CI still turned off. Seven children with normal hearing formed a comparison group. They were also tested twice, with a 1-hr intermediate interval. IOP and IOP duration were measured for the medial consonant in both auditory conditions. The results show that auditory condition affected peak IOP more so than IOP duration. Peak IOP was greater for /p/ than /b/ with the CI off, but some participants reduced or reversed this contrast when the CI was on. The findings suggest that different speakers with CIs may use different speech production strategies as they learn to use the auditory signal for speech

HYPOTHESES: Do cochlear implants provide enough information to allow adult cochlear implant users to understand words in ways that are similar to listeners with acoustic hearing? Can we use a computational model to gain insight into the underlying mechanisms used by cochlear implant users to recognize spoken words? BACKGROUND: The Neighborhood Activation Model has been shown to be a reasonable model of word recognition for listeners with normal hearing. The Neighborhood Activation Model assumes that words are recognized in relation to other similar-sounding words in a listener's lexicon. The probability of correctly identifying a word is based on the phoneme perception probabilities from a listener's closed-set consonant and vowel confusion matrices modified by the relative frequency of occurrence of the target word compared with similar-sounding words (neighbors). Common words with few similar-sounding neighbors are more likely to be selected as responses than less common words with many similar-sounding neighbors. Recent studies have shown that several of the assumptions of the Neighborhood Activation Model also hold true for cochlear implant users. METHODS: Closed-set consonant and vowel confusion matrices were obtained from 26 postlingually deafened adults who use cochlear implants. Confusion matrices were used to represent input errors to the Neighborhood Activation Model. Responses to the different stimuli were then generated by the Neighborhood Activation Model after incorporating the frequency of occurrence counts of the stimuli and their neighbors. Model outputs were compared with obtained performance measures on the Consonant-Vowel Nucleus-Consonant word test. Information transmission analysis was used to assess whether the Neighborhood Activation Model was able to successfully generate and predict word and individual phoneme recognition by cochlear implant users. RESULTS: The Neighborhood Activation Model predicted Consonant-Vowel Nucleus-Consonant test words at levels similar to those correctly identified by the cochlear implant users. The Neighborhood Activation Model also predicted phoneme feature information well. CONCLUSION: The results obtained suggest that the Neighborhood Activation Model provides a reasonable explanation of word recognition by postlingually deafened adults after cochlear implantation. It appears that multichannel cochlear implants give cochlear implant users access to their mental lexicons in a manner that is similar to listeners with acoustic hearing. The lexical properties of the test stimuli used to assess performance are important to spoken-word recognition and should be included in further models of the word recognition process

A method is presented for analyzing cochlear implant stimulations and typical representations used in simulations. Filtered "white-noise" bands are modulated using sinusoids, representing differing stimulation channels. These representations, along with their corresponding envelopes, are used to generate neural activation patterns (NAPs), which represent "normal-hearing" responses in the auditory nerve to these stimuli. Additionally, NAPs are generated to represent the neural activity induced by cochlear implant stimulation strategies, assuming exponential rolloff from the electrodes. The mean squared error is measured between NAPs both directly, and after compensation for perceptual resolution. Results suggest that the noise-band approximation of the CIS implant signal actually has more in common with the original source than with the implant stimulation patterns.

Children with profound congenital or prelingual deafness encounter significant difficulties in the development of skills in an oral language such as English. Their language development, however, can be accelerated if they receive a cochlear implant-a sensory aid that facilitates language acquisition by providing important auditory information. The present study used the Reynell Developmental Language Scales (RDLS) to assess language skills pre- and postimplant in 44 pediatric cochlear implant users. All users were profoundly to totally deaf, either at birth or before the age of 3 years. They all received cochlear implants before the age of 6 and were programmed with state-of-the-art stimulation strategies (CIS or SPEAK) since the day of initial stimulation. The main finding was that postimplantation language development proceeded at a pace that was not significantly different from normal. Thus, the language gap present at implantation did not increase after children started using the device, as it would if they had not received cochlear implants. Nevertheless, it is important to conduct further studies to determine whether these conclusions apply when other language skills, such as the use of grammar, are measured

The goal of this study was to test 2 hypotheses about language development in pediatric cochlear implant (CI) users. The 'language instinct' hypothesis states that children with CIs will develop language in the same sequence as children with normal hearing, but in a delayed fashion. In other words, noun plurals will develop first, and the use of the uncontractible copula and regular past tense will follow. An alternative hypothesis (the 'perceptual prominence' hypothesis) is that the pattern of language development in CI users will be strongly affected by the perceptual prominence of the relevant morphological markers. This hypothesis predicts that the uncontractible copula will develop first, followed by noun plurals, and then by regular past tense. A sentence completion task was used to measure the performance of 9 pediatric CI users and compare it to that of several groups of children with normal hearing. The results from the CI users were consistent with the perceptual prominence hypothesis. In particular, the scores for the copula probe were higher than those for the noun plural probe for 8 of the 9 CI users. This result represents a rather striking inversion with respect to the usual development pattern in children with normal hearing and even in children with specific language impairment. If the perceptual prominence hypothesis receives further support in future studies, clinicians who work in language rehabilitation of CI users may choose to target those aspects of grammar that are less acoustically prominent to these children. In addition, and from a theoretical standpoint, these results suggest that although there may well be an innate language acquisition mechanism, patterns of language development can be strongly affected by the acoustic input. $$: