Optimization of speech processor fitting strategies for Chinese-speaking cochlear implantees

Comparison of Mandarin tone and speech perception between advanced combination encoder and continuous interleaved sampling speech-processing strategies in children

OBJECTIVE

This study was performed to compare cochlear implant (CI) users' performance in Mandarin speech and tone perception between 2 types of speech-processing strategies-advanced combination encoder (ACE) and continuous interleaved sampling (CIS)-under quiet and noisy conditions.

METHODS

This study involved 10 congenitally deaf children (age range, 5.7-15.3 years; mean, 9.2 years) who received the Nucleus 24-channel CI system cochlear device (CI24R; Cochlear Ltd, Lane Cove NSW, Australia). The subjects used ACE since switching on their CI devices. Speech and tone perception tests were administered under quiet and noisy (+5 dB signal-to-noise ratio) conditions with ACE and CIS strategies 20 minutes and 2 weeks apart.

RESULTS

Regardless of the strategy used, subjects showed significantly higher scores in speech perception than in tone recognition. Under noisy conditions, subjects had significantly higher tone identification scores with the CIS than the ACE strategy (P = .038). There was no significant difference in speech identification score between the strategies. Subjects showed significant higher tone identification and speech perception scores under quiet than noisy (+5 dB signal-to-noise ratio) conditions. Subjectively, 6 subjects preferred the ACE strategy, and the remaining 4 preferred the CIS strategy. The strategy preference of the subjects was related to speech perception performance rather than tone identification. A significant correlation was observed between tone identification and speech recognition, regardless of whether speech was evaluated by consonants (r = 0.669, P < .001), vowels (r = 0.426, P = .001), or sentences (r = 0.294, P = .023).

CONCLUSION

There are only 4 patterns of tone in Mandarin, which is far fewer than the number of speech sounds. However, tone identification is poorer than speech perception. The CIS speech-processing strategy may improve tone identification under noisy conditions. Before improved speech strategies to code acoustic characteristics of tone can be developed, it would be worthwhile to try both CIS and ACE for CI users and to select the most suitable speech-processing strategy according to the subjective preference and objective performance.

Lexical tone perception with HiResolution and HiResolution 120 sound-processing strategies in pediatric Mandarin-speaking cochlear implant users

OBJECTIVES

Lexical tone recognition tends to be poor in cochlear implant users. The HiResolution (HiRes) sound-processing strategy is designed to better preserve temporal fine structure, or the detailed envelope information, of an acoustic signal. The newer HiRes 120 strategy builds on HiRes by increasing the amount of potential spectral information delivered to the implant user. The purpose of this study was to examine lexical tone recognition in native Mandarin Chinese-speaking children with cochlear implants using the HiRes and HiRes 120 sound-processing strategies. Tone recognition performance was tested with HiRes at baseline and then after up to 6 mo of HiRes 120 experience in the same subjects.

DESIGN

Twenty prelingually deafened, native Mandarin-speaking children, with ages ranging from 3.5 to 16.5 yr, participated. All children completed a computerized tone contrast test on three occasions: (1) using HiRes immediately before conversion to HiRes 120 (baseline), (2) 1 mo after conversion, and (3) 3 mo after conversion. Twelve of the 20 children also were tested 6 mo after conversion. In addition, the parents of 18 children completed a questionnaire at the 3-mo follow-up visit regarding the preference of sound-processing strategies and the children's experience related to various aspects of auditory perception and speech production using HiRes 120.

RESULTS

As a group, no statistically significant differences were seen between the tone recognition scores using HiRes and HiRes 120. Individual scores showed great variability. Tone recognition performance ranged from chance (50% correct) to nearly perfect. Using the conventional HiRes strategy, 6 of the 20 children achieved high-level tone recognition performance (i.e., >or=90% correct), whereas 7 performed at a level not significantly different from chance (50-60% correct). At the final test, either 3 or 6 mo after conversion, all children achieved tone recognition performance with HiRes 120 that was equal to or better than that with HiRes, although some children's tone recognition performance was worse initially at the 1 or 3 mo follow-up intervals than at baseline. Eight of the 20 children showed statistically significant improvement in tone recognition performance with HiRes 120 on at least one of the follow-up tests. Age at implantation was correlated with tone recognition performance at all four test intervals. Parents of most of the children indicated that the children preferred HiRes 120 more than HiRes.

CONCLUSIONS

As a group, HiRes 120 did not provide significantly improved lexical tone recognition compared to HiRes, at least throughout the length of the study (up to 6 mo). There were large individual differences in lexical tone recognition among the prelingually deafened, native Mandarin-speaking children with cochlear implants using either HiRes or HiRes 120. Six of the 20 children performed at or near ceiling in the baseline HiRes condition. Of the remainder, approximately half showed significantly better tone recognition when subsequently tested with HiRes 120, although the extent to which this improvement may be attributable to factors other than the change in processing strategy (e.g., general development) is unknown. The children who benefited most from HiRes 120 tended to be those who were implanted at younger ages.

Lexical tone recognition with an artificial neural network

OBJECTIVES

Tone production is particularly important for communicating in tone languages such as Mandarin Chinese. In the present study, an artificial neural network was used to recognize tones produced by adult native speakers. The purposes of the study were (1) to test the sensitivity of the neural network to speaker variation typically in adult speaker groups, (2) to evaluate two normalization procedures to overcome the effects of speaker variation, and (3) to compare tone recognition performance of the neural network with that of the human listeners.

DESIGN

A feedforward multilayer neural network was used. Twenty-nine adult native Mandarin Chinese speakers were recruited to record tone samples. The F0 contours of the vowel part of the 1044 monosyllabic words recorded were extracted using an autocorrelation method. Samples from the F0 contours were used as inputs to the neural network. The efficacy of the neural network was first tested by varying the number of inputs and the number of neurons in the hidden layer from 1 to 16. The sensitivity of the neural network to speaker variation was tested by (1) using the raw F0 data from speech tokens of a number of randomly drawn speakers that varied from 1 to 29, (2) using the raw F0 data from speech tokens of either male-only or female-only speakers, and (3) using two sets of normalized F0 data (i.e., tone 1-based normalization and first-order derivative) from speech tokens from a number of randomly drawn speakers that varied from 1 to 29. The recognition performance of the neural network under several experimental conditions was compared with the corresponding recognition performance of 10 normal-hearing, native Mandarin Chinese speaking adult listeners.

RESULTS

Three inputs and four hidden neurons were found to be sufficient for the neural network to perform at about 85% correct using speech samples without normalization. The performance of the neural network was affected by variation across speakers particularly between genders. Using the tone 1-based normalization procedure, the performance of the neural network improved significantly. The recognition accuracy of the neural network as a whole or for each tone was comparable with that of the human listeners.

CONCLUSIONS

The neural network can be used to evaluate the tone production of Mandarin Chinese speaking adults with human listener-like recognition accuracy. The tone 1-based normalization procedure improves the performance of the neural network to human listener-like accuracy. The success of our neural network in recognizing tones from multiple speakers supports its utility for evaluating tone production. Further testing of the neural network with hearing-impaired speakers might reveal its potential use for clinical evaluation of tone production.

Spectral and temporal cues for speech recognition: implications for auditory prostheses

Features of stimulation important for speech recognition in people with normal hearing and in people using implanted auditory prostheses include spectral information represented by place of stimulation along the tonotopic axis and temporal information represented in low-frequency envelopes of the signal. The relative contributions of these features to speech recognition and their interactions have been studied using vocoder-like simulations of cochlear implant speech processors presented to listeners with normal hearing. In these studies, spectral/place information was manipulated by varying the number of channels and the temporal-envelope information was manipulated by varying the lowpass cutoffs of the envelope extractors. Consonant and vowel recognition in quiet reached plateau at 8 and 12 channels and lowpass cutoff frequencies of 16 Hz and 4 Hz, respectively. Phoneme (especially vowel) recognition in noise required larger numbers of channels. Lexical tone recognition required larger numbers of channels and higher lowpass cutoff frequencies. There was a tradeoff between spectral/place and temporal-envelope requirements. Most current auditory prostheses seem to deliver adequate temporal-envelope information, but the number of effective channels is suboptimal, particularly for speech recognition in noise, lexical tone recognition, and music perception.

Relation Between Neural Response Telemetry Thresholds, T- and C-Levels, and Loudness Judgments in 12 Adult Nucleus 24 Cochlear Implant Recipients

OBJECTIVE

The primary purpose of this study was to determine if the contour of visual (vNRT) or predicted (tNRT) neural response telemetry (NRT) thresholds across electrodes could predict the contour of behaviorally programmed T-levels (minimum stimulation) and/or C-levels (maximum stimulation) across electrodes for well-fit MAPs. The secondary purpose was to determine the relation between NRT thresholds and loudness judgments obtained at the subject's MAP rate (250, 900, 1200, or 1800 pulses per second [pps]) and the NRT stimulus rate (80 pps).

DESIGN

Twelve adult Nucleus 24 cochlear implant recipients participated in the study. The T- and C-levels from a preferred MAP, which had been worn for a minimum of 3 mo, were used in this study. Electrically evoked compound action potentials were measured on 11 active electrodes with NRT software (v3.0). Ascending loudness judgments from first hearing to maximum acceptable loudness were completed on these electrodes with the subject's preferred MAP rate stimulus, using the R126 (v.2.0) software and with an 80 pps rate stimulus, using the NRT software (v3.0). All measures were repeated approximately 1 mo later to determine their reliability.

RESULTS

The reliability of the behavioral and objective measures was very high from the first to the second half of the study. The mean tNRT thresholds had a lower reliability (r = 0.73) than vNRT thresholds (r = 0.91). The loudness judgment dynamic range was notably different between rates. The NRT rate (80 pps) stimulus resulted in the narrowest dynamic range followed by increasingly wider dynamic range as the MAP rate increased. The NRT thresholds had a stronger correlation with loudness judgments made with the NRT rate stimulus than with the MAP rate stimulus. The group mean NRT thresholds were significantly correlated with C-levels (vNRT r = 0.69) (tNRT r = 0.66) but not T-levels. The relation between NRT thresholds and T- and C-levels varied for different MAP rates, with the NRT thresholds being closest to the C-levels for the 250 pps MAP rate. Each subject's vNRT thresholds and MAP levels were examined by fitting a third-order polynomial to the data. This analysis revealed significant variability demonstrating that no one fit predicts T- and C-levels well for all subjects.

CONCLUSIONS

The results of this study provide important insight into the relation between NRT thresholds and loudness judgments for different stimulation rates and T- and C-levels at various MAP rates. The loudness judgment dynamic range and MAP dynamic range (T- and C-levels) varied notably for different stimulation rates. As a result, the relation of NRT thresholds to these measures also varied with stimulation rate. Overall, the mean vNRT thresholds fell higher in the loudness judgment dynamic range than the tNRT thresholds. Mean NRT thresholds fell between the judgments of medium soft and maximum acceptable loudness for all stimulation rates. Mean vNRT thresholds fell above C-levels, whereas almost half of tNRT thresholds fell just below C-levels. However, the relation between NRT thresholds and C-levels varied substantially for different MAP stimulation levels. In addition, there is substantial individual variability in the relation between NRT thresholds and MAP levels that is not reflected in the group data. The prediction of the contour of T- and C-levels from the contour of NRT thresholds across electrodes would not be appropriate for half of the subjects. Therefore, great care should be taken when applying a fitting rule that incorporates NRT thresholds without considering these individual differences. For adults who can provide appropriate loudness judgments and threshold responses it appears to be most efficient to primarily use behavioral measures to create MAPs.

Effects of limiting the number of active electrodes on Mandarin tone perception in young children using cochlear implants

OBJECTIVE

To evaluate the short-term effects of varying the number and location of programmed electrodes on Mandarin tone perception in young children using Nucleus CI 24 implants.

MATERIAL AND METHODS

Six children using CI 24 implants participated in the study. The Mandarin tone recognition of each subject was assessed using a self-designed word list containing 80 words under 5 different map conditions. The map conditions were: (i) the subject's current map with all functional electrodes; (ii) removing (i.e. not programming) the even-numbered active electrodes; (iii) removing the five most apical electrodes; (iv) removing the six most basal electrodes; and (v) removing all electrodes except the six most apical.

RESULTS

Reducing the number of active electrodes caused a small but significant decrease in performance of Mandarin tone perception. However, a relatively high tone perception score (58.1%) could be maintained even when subjects used only six apical electrodes. The location of the electrodes, i.e. apical or basal, did not appear to be important in tone perception. For each individual tone, the scores for tones 1 and 4 were significantly higher than those for tones 2 and 3 for all map conditions. Comparing tone perception with word recognition, reducing the number of active electrodes had a much more negative effect on word recognition.

CONCLUSIONS

A small but significant decrease in Mandarin tone perception was observed when the number of active electrodes was reduced in children using CI 24 implants. The location of the electrode is not important in tone recognition. Moreover, tone recognition is far more resistant than word recognition to the negative effects of reducing the number of active programmable electrodes, possibly because it relies mostly on temporal envelopes while word recognition requires more spectral details.

Tone production in Mandarin-speaking children with cochlear implants: a preliminary study

OBJECTIVE

More than a quarter of the world's population speak tone languages, such as Mandarin Chinese. In those languages, the pitch or tone pattern of a monosyllabic word conveys lexical meaning. The purpose of this study was to investigate tone production in Mandarin-speaking children with cochlear implants (CIs).

MATERIAL AND METHODS

Speech samples were recorded from seven normal-hearing and four CI children aged 4-9 years. All subjects were native Mandarin speakers. The speech samples were used for acoustic analysis of the tone patterns, i.e. the fundamental frequency contours. In addition, a tone intelligibility test was carried out in which four normal-hearing native Mandarin-speaking adults listened to the speech materials and judged the intelligibility of the children's tone production.

RESULTS

The tone production for the seven normal-hearing children was considered to be perfect in the intelligibility test. Acoustic analysis of the speech materials of the normal-hearing children produced the four typical tone patterns of Mandarin Chinese: (i) high and flat; (ii) rising; (iii) low and dipping; and (iv) falling. The tone patterns produced by the children with CIs tended to be flat, with some other patterns being irregular. The results of the tone intelligibility tests also showed degraded intelligibility of tone patterns.

CONCLUSION

A potential speech development deficit was documented in prelingually deafened children with CIs whose native language is a tone language. The imperfect tone production of the implant children, which can be attributed to the paucity of pitch information delivered via the current CI stimulation, may have significant implications for communication using tone languages. Further research is warranted to determine factors that may affect tone development in children with CIs.

Effects of presentation level on phoneme and sentence recognition in quiet by cochlear implant listeners

OBJECTIVE

The objectives of this study were to characterize the effects of presentation level on speech recognition in quiet by cochlear implant users with the Nucleus 22 SPEAK and Clarion v1.2 CIS speech-processing strategies, and to relate speech recognition at low presentation levels to stimulus audibility as measured by sound field thresholds. It was hypothesized that speech recognition performance in both Nucleus SPEAK and Clarion CIS participants would decrease as presentation level was decreased below 50 to 60 dBA, due to audibility limitations. However, it was expected that such level effects would be less severe in CIS participants than in SPEAK participants because the Clarion v1.2 device encodes a wider acoustic dynamic range (up to 60 dB) than the Nucleus 22 device (30 dB).

DESIGN

Performance-intensity (P-I) functions for vowels, consonants and sentences in quiet were obtained from each participant. P-I functions incorporated speech levels of 70, 60, 50, 40 and 30 dBA. Subjects used their clinical speech processor maps and adjusted the loudness (volume/sensitivity) controls on their processors so that speech presented at 60 dBA was comfortably loud. Maps were created using default clinical procedures and were not adjusted to optimize sound field thresholds. Sound field thresholds and dynamic ranges were measured for warbled pure tones with frequencies of 250 to 6000 Hz.

RESULTS

Consonant and sentence recognition showed strong level effects for both SPEAK and CIS participants, with performance decreasing substantially at levels below 50 dBA in most individuals. Vowel recognition showed weaker level effects. For all three speech materials, SPEAK and CIS participants demonstrated similar mean performance at 70 dBA; however, SPEAK participants showed larger reductions in performance than CIS participants with decreasing level. Sound field thresholds were more sensitive for CIS participants than for SPEAK participants, supporting the hypothesis that performance differences were related to audibility.

CONCLUSIONS

Cochlear implant listeners are unable to maintain good speech recognition at low presentation levels due to reduced stimulus audibility, and this may significantly limit their ability to communicate in daily life. It is likely that audibility differences between SPEAK and CIS participants in the present study can be attributed at least partly to differences in the acoustic dynamic range used by the respective processors. However, several additional factors may have contributed to differences in audibility and perception of soft speech among individual listeners with both devices. These include the minimum and maximum electrical stimulation levels specified in participants' maps and the speech processor sensitivity setting used for testing.

Features of stimulation affecting tonal-speech perception: implications for cochlear prostheses

Tone languages differ from English in that the pitch pattern of a single-syllable word conveys lexical meaning. In the present study, dependence of tonal-speech perception on features of the stimulation was examined using an acoustic simulation of a CIS-type speech-processing strategy for cochlear prostheses. Contributions of spectral features of the speech signals were assessed by varying the number of filter bands, while contributions of temporal envelope features were assessed by varying the low-pass cutoff frequency used for extracting the amplitude envelopes. Ten normal-hearing native Mandarin Chinese speakers were tested. When the low-pass cutoff frequency was fixed at 512 Hz, consonant, vowel, and sentence recognition improved as a function of the number of channels and reached plateau at 4 to 6 channels. Subjective judgments of sound quality continued to improve as the number of channels increased to 12, the highest number tested. Tone recognition, i.e., recognition of the four Mandarin tone patterns, depended on both the number of channels and the low-pass cutoff frequency. The trade-off between the temporal and spectral cues for tone recognition indicates that temporal cues can compensate for diminished spectral cues for tone recognition and vice versa. An additional tone recognition experiment using syllables of equal duration showed a marked decrease in performance, indicating that duration cues contribute to tone recognition. A third experiment showed that recognition of processed FM patterns that mimic Mandarin tone patterns was poor when temporal envelope and duration cues were removed.

Effect of speech processor program modifications on cochlear implant recipients' threshold and maximum acceptable loudness levels

This study's purpose was to determine whether or not modifications in speech processor electrical stimulation levels were associated with changes in five Nucleus 22 cochlear implant recipients' thresholds or maximum acceptable loudness levels (MALs). These modifications in minimum and maximum stimulation levels were made to optimize hearing in everyday life. One threshold and one MAL were obtained on each active electrode during six, weekly test sessions, three before and three after program modification. Only one participant had a significant change in threshold after program modification; this participant and four others had significant changes in MAL. Participants' threshold variability was the same, but MAL variability was higher than that observed in other studies. Because these participants had no experience making MAL judgments prior to this study, this result suggests that implant recipients should be given sufficient practice in making MAL judgments to provide a stable clinical estimate of the upper boundary of the electrical dynamic range.