Frequency-limiting effect on perception of speech and environmental sound in normal-hearing and cochlear implant adults

Abstract

Background: Spectral information plays an important role for speech perception as well as environmental sounds. A cochlear implant (CI) can provide extended spectral information compared to conventional hearing aids resulting in improved speech perception by the implantee. However, although the speech sounds depend upon the language of the target speech, CI has been developed and researched primarily in English. The languages around world contain language-universal characteristics as well as language-specific characteristics of perception and production. Korean language is different from English in many aspects, so it might affect the perception performance of CI speech processing mainly developed with English language. It would be worthy to investigate how different frequency-limiting conditions produced by high-pass and low-pass filtering affect sound recognition in normal-hearing (NH) and CI subjects to examine the effect of frequency bands on speech in Korean and environmental sound perception. Objectives: The purposes of this study are first, to find if there is perceptual differences between NH and CI subjects in speech and environmental sounds according to the frequency-limiting conditions, second, how importance of frequency information shows in Korean NH and CI listeners, and third, how the individual phonemes and sounds are recognized in a variety of frequency-limiting conditions. Methods: We conducted vowel and consonant tests in Korean (pronounced by one male and one female speakers) and an environmental sound test for 10 NH and 10 CI subjects in quiet within a variety of frequency-limiting conditions. We used the original sound and frequency band from 100 Hz to 8000 Hz. The band of 100 – 8000 Hz was low-pass and high-pass filtered to derive 28 test conditions for each phoneme and 14 test conditions for each environmental sound per subject. Results: We found a significant interaction effect on sound recognition according to the group, frequency bands, and sound stimuli. There were similarities and differences in recognition pattern between NH and CI group. As for the similarities, first, both group showed better sound recognition as more frequency region was added. Second, as more frequency information introduced, there was more increase of vowel and environmental sound recognition than that of consonant recognition. Third, both groups did not show recognition difference between original sound and 100 – 8000 Hz full band in all the stimuli. Fourth, in the original sound and full band conditions, both groups showed recognition scores as vowel > environmental sound > consonant, in order of performance. Fifth, environmental sound recognition was better than speech recognition in low- and high-pass filtering condition in both groups. Sixth, crossover frequency was occurred around 1500 Hz in all the sound stimuli in both groups. Seventh, in environmental sounds, both groups showed frequency redundancy in almost all the frequency bands. Eighth, on average, both groups recognized vowel /i/ and consonant /h/ best. Ninth, environmental sounds could be divided into the group of sounds which were easily recognized with low frequency or with high frequency and both groups recognized those sounds in a similar pattern. As for dissimilarities, first, CI group got more benefits in vowel recognition than in consonant recognition as frequency information increased and most easily recognized environmental sounds, then vowels and consonants in order. Second, NH group showed difficulty of recognizing phonemes compare to environmental sounds in high-pass filtering whereas CI group showed difficulty of recognizing consonants compare to vowels and environmental sounds in full band + original sound condition. Third, CI group did not utilize frequency cues for consonant recognition as efficient as NH group. Fourth, CI results did not demonstrate perceptual saturation but rather linear increase as more high frequency introduced. Fifth, when we compared full band recognition scores to summation of low-pass and high-pass filtering scores, the NH group showed significant redundancy of frequency at 1500 Hz in vowel and all the cutoff frequency in consonant whereas the CI group did not show redundancy in any cutoff frequency which resulted in no saturation pattern and linearity in recognition with CI speech processing. Sixth, gradient effects confirmed that there were more low frequency information in low-pass filtering and more high frequency information in high-pass filtering for the NH group whereas a reduced and even distribution of frequency information was seen for the CI group across frequency bands. Seventh, CI subjects showed larger individual difference in environmental sound recognition scores, but not in speech sound recognition. Eighth, both phoneme and sound recognition changed according to the frequency bands, and in general, both groups showed a recognition pattern change along formants of vowels; however, the CI group showed varied consonant recognition in a variety of frequency-limiting conditions more than NH group. Ninth, in general, CI group utilized frequency information above 2500 Hz for vowel and consonant recognition and frontal vowels, fricatives, affricates, and alveolars were benefited by it in spite of reduced amount of information about phoneme perception. Discussion: Both groups reflected several perceptual similarities with regard to acoustical characteristics. However, CI group demonstrated different pattern of sound recognition and had difficulty in recognizing different types of sounds with regard to the frequency-limiting conditions. This implies that speech and environmental sound recognition were affected differently from NH group by frequency utilization ability of CI subjects. It seems that 1500Hz brought a significant recognition change in speech for normal Korean listeners, which reflects a speech emphasis in the lower frequency domain in Korean language than English. The CI group also showed crossover around 1500 Hz which show that they recognized specific frequency characteristics of Korean language. With the results that CI users showed linear improvement in sound recognition with increased frequency information. Especially recognition of fricatives, affricates, and alveolars was gradually increased up to 8000 Hz so that we concluded that CI group in this study could utilize frequency information from 100 Hz to 8000 Hz provided by Current CI speech processing and the cues in frequency information conveyed by the device played perceptual role for them. However, we should not overlook the individual differences in recognition according to the frequency bands and individual phonemes and sounds. It is of interest that environmental sound also showed 1500 Hz crossover. On the other hand, environmental sound was revealed to have similar frequency pattern to speech sound in Korean listener and it would be interesting to see if English listeners show the same results to this study. Besides, other factors than the frequency factor in this study seemed to act on the recognition in addition to frequency information in both groups of the subjects. Conclusion: Korean CI listeners utilized 100 – 8000 Hz frequency information to recognize speech and environmental sounds. However, Korean has around 1500 Hz magnitude in speech perception and CI subjects showed different recognition patterns of individual phonemes and sounds with regard to frequency-limiting conditions. With these results, we concluded that CI devices of Koreans users need to be programmed differently and individually to maximize utilization of frequency information in 100 – 8000 Hz as the CI users showed different phoneme and environmental sound recognition according to frequency-limiting conditions. These findings potentially have predictive implications and can be utilized to customize the programming and rehabilitation for individual CI users who use Korean language.