Envelope Versus Fine Structure Speech Coding Strategy: A Crossover Study

Transcription

1 Otology & Neurotology 32:1094Y1101 Ó 2011, Otology & Neurotology, Inc. Envelope Versus Fine Structure Speech Coding Strategy: A Crossover Study *Dominik Riss, *Jafar-Sasan Hamzavi, *Andreas Selberherr, Alexandra Kaider, *Michaela Blineder, *Veronika Starlinger, *Wolfgang Gstoettner, and *Christoph Arnoldner *Department of Otorhinolaryngology, Medical University of Vienna; and ÞCenter for Medical Statistics, Informatics and Intelligent Systems, Section for Clinical Biometrics, Medical University of Vienna, Vienna, Austria Objective: The transmission of fine structure information to cochlear implant users is an expanding area of research. Previous studies comparing the fine structure processing (FSP) speech coding strategy to the envelope-based continuous interleaved sampling (CIS) strategy indicated improved speech perception when using the fine structure strategy. Those investigations were performed with an extended frequency spectrum in the low frequencies together with the fine structure strategy. The current study addresses the question whether these improvements are due to the presentation of fine structure per se or rather the extended frequency spectrum. Hence, this cross over study compares the two strategies using an identical frequency spectrum. Study Design: Randomized crossover study. Patients: 31 patients were randomly assigned to two groups. Interventions: One group was fitted with a CIS map for 4 weeks, tested and subsequently fitted with a FSP map for 4 weeks. The other group followed the same pattern in reverse. Main Outcome Measures: Test material consisted of sentence tests in noise, monosyllables in quiet and melody recognition. Results: No statistical significance was noted between the different speech coding strategies at an identical frequency spectrum. Conclusion: This study shows that there is no difference in speech perception with FSP compared to CIS at an extended frequency spectrum. Therefore, the extended frequency spectrum in the low frequencies might explain a benefit of FSP observed in previous studies. Key Words: Cochlear implantvfrequency rangev Music perceptionvspeech coding strategyvspeech perception. Otol Neurotol 32:1094Y1101, Address correspondence and reprint requests to Christoph Arnoldner, M.D., ENT Department AKH-HNO, Medical University of Vienna, Waehringer Guertel 18-20, A-1097 Vienna, Austria; christoph.arnoldner@ meduniwien.ac.at Sources of support: This academic study was conducted at the Medical University of Vienna. Technical support was provided by the cochlear implant company MED-EL (Innsbruck, Austria), in particular Dominik Richnovsky. One of the authors (W. G.) is a recipient of research grants by MED-EL. Over time, speech perception with cochlear implants has drastically improved to a point where patients reach near-normal levels in quiet environments. The algorithms to translate the acoustic signal into patterns of electrical stimulationvthe so-called speech coding strategiesvare one of the factors that have helped to achieve these high levels of speech perception. However, there is still need for improvement when it comes to the ability to understand speech in noise as well as music perception. One widely used strategy is continuous interleaved sampling (CIS) (1). The acoustic signal is band-pass filtered using one filter for each electrode, with the envelope function calculated for each filter output. In every cycle of stimulation, each electrode is stimulated in a specific interleaved sequence, which achieves the greatest possible distance between the subsequently activated electrodes. In this way, electrical field overlapsv so-called channel interactionsvare reduced. The timing of each pulse is set according to the stimulation rate. Therefore, no information is transmitted by the timing of the electric pulse. Because the amplitude of each pulse is set according to the envelope function, this strategy often is described as envelope based (2). According to Hilbert (3), an acoustic signal can be represented by an envelope function modulating a highfrequency carrier. Since Smith et al. (4) have demonstrated the importance of fine structure for pitch perception and sound localization, several attempts have been made to increase the representation of fine structure through new coding strategies. To improve the spectral resolution and hence aim at spectral fine structure, current steering 1094

2 ENVELOPE VERSUS FINE STRUCTURE 1095 techniques have been developed (5Y7). These create virtual channels by adjusting currents with the idea of increasing the number of independent effective channels. On the other hand, considerable effort has been invested in improving temporal fine structure representation (8). A study by Zeng et al. (9) investigated the role of amplitude and frequency modulation in realistic listening situations. It was found that frequency modulation provides independent and complementary contributions compared with amplitude modulation in these conditions. In an effort to improve signal processing accordingly, frequency modulation was extracted from temporal fine structure, and it was used to frequency modulate the carrier (10). A different possibility was proposed by Throckmorton et al. (11). They used multiple carrier frequencies per channel to encode fine structure. In a third way to encode fine structure, one company (MED-EL, Innsbruck, Austria) has introduced the fine structure processing (FSP) speech coding strategy (12). This is a CIS-based strategy in which the timing of stimulation on a number of the most apical electrodes is determined by the zero crossing of the band-pass filter output (13). At that moment, a pulse train is started with the amplitude according to the envelope function (Fig. 1A). The FIG. 1. Differences between CIS and FSP. A, This scheme shows a band-pass filter output (top) with the continuous line representing the original filtered signal and the dotted line representing the envelope function. FSP (middle) stimulates at the zero crossing of the bandpass output with an amplitude according to the envelope function. CIS (bottom) stimulates at a fixed stimulation rate according to the envelope function. B, This shows an example of stimulation pulses comparing FSP on the left and CIS on the right. Channels 3 to 12 present envelope information (i.e., amplitude fluctuations) carried by a constant, high rate pulse train. In Channels 1 and 2 with FSP (left), the timing of stimulation reflects the frequency of the original signal, whereas in CIS (right), the timing of stimulation is set according to the rate. (Both figures reprinted with permission from MED-EL, Innsbruck Austria.)

3 1096 D. RISS ET AL. higher the per channel stimulation rate, the more apical channels can be stimulated in this manner (maximum of 3). Frequency variations within channels are therefore possibly transmitted in the low-frequency spectrum. All other channels are stimulated according to the CIS strategy. The difference between CIS and FSP can be found in the transmission of the acoustic signal in the low frequencies up to about 300 to 470 Hz, which is represented by 2 or 3 most apical channels. Whereas the timing of stimulation is set solely by the stimulation rate in CIS, FSP adapts the timing of stimulation according to the frequency of the original signal in this frequency band. Thereby, temporal information is introduced and presented to the user. This aims at an improved transmission of within channel frequency variations and thereby more temporal fine structure is possibly presented to the user. Stimulation patterns of Electrodes 4 to 12 are identical for both coding strategies. A graphic example of stimulation patterns between coding strategies is presented in Figure 1B. The transmission of fine structure in the low frequencies is especially important for the transmission of the fundamental frequency (F0) of speech sounds. F0 information has been shown to be beneficial for identifying the speaker and thereby enhance speech understanding in noise (14). Previous studies comparing the 2 speech coding strategies (15,16) noted a significant improvement in speech perception especially in difficult listening situations with FSP compared with CIS. These studies were conducted when patients were upgraded from the Tempo+, which uses CIS, to the Opus speech processor, which offers FSP. In addition, the Opus processor also offers an extended frequency spectrum in the low frequencies down to 70 Hz, whereas the Tempo+ has a lower cutoff frequency of 200 Hz. Therefore, it cannot be determined whether the reported advantage was in fact due to the speech coding strategy or rather due to the extended frequency range. FIG. 2. Crossover study design. After a baseline test, 31 patients were randomly assigned to 2 groups and received either FSP for 4 weeks and then CIS for 4 weeks or vice versa. At each visit, a sentence test in noise, a monosyllables test, and a music test were performed. A within-subject statistical analysis was performed using an ANCOVA model to account for the crossover design. Baseline values (Visit 1) were only used as a covariate in the statistical model. The primary aim of this study was to compare the 2 speech coding strategies, FSP and CIS, at an identical, extended frequency spectrum down to 70 Hz. A crossover study design was used to minimize learning effects and allow within-subject comparison (Fig. 2). The primary end point of this study was sentence recognition in noise. Secondary end points were monosyllabic word scores in quiet as well as music perception. MATERIALS AND METHODS Patients Thirty-one postlingually deaf patients (Table 1), with a minimum experience of 1 year with their cochlear implant, were enrolled in the study. All of them used the Opus speech processor (MED-EL, Innsbruck, Austria), which uses FSP. The mean patient age was 60.9 years (range, 40.1Y79.4). Two patients were tested with bilateral cochlear implants (patients 20 and 25). They had a minimum of 1 year experience with their Opus speech processor using FSP on both ears. These 2 patients were tested with both implants activated and programmed to the same speech coding strategy, according to study protocol. Additional demographic details are listed in Table 1.The study was approved by the ethics committee of the Medical University of Vienna. Frequency Range For both speech coding strategies, an identical, extended frequency spectrum starting at 70 Hz was used. Individual corner frequencies for each frequency band are displayed in Table 2. Frequencies for FSP and CIS did not differ from the FSP map that patients had previous experience with. For CIS settings, the CIS implementation of the OPUS speech processor was used, which is also called HDCIS. For simplicity reasons, this implementation is referred to as CIS throughout this article. Crossover Study Design All patients were already using FSP at the beginning of the study. At the first visit, they were tested with their familiar map to obtain baseline scores. Patients were randomly assigned to one of the 2 groups: group A was fitted with a CIS map and was instructed to use it for 4 weeks. Group B received an FSP map. At the second visit, 4 weeks later, all patients were tested and then switched to the alternative speech coding strategy: group A received FSP, whereas group B received CIS. Patients were again instructed to use their maps for 4 weeks, and the third and last visit was scheduled thereafter. At this last visit, the patients were tested, the study was completed and the subjects were set to their preferred map. A graphic scheme of the study is presented in Figure 2. All maps were fitted individually to achieve the best possible hearing sensation for each patient. Per channel stimulation rates were kept at the same level when switching speech coding strategies. FSP maps were similar to the ones used before the beginning of the study, only with slight adjustments, which are customary at a regular control visit to optimize loudness and sound quality. At each visit, sentences in noise, monosyllables in quiet, and music perception were tested. The results of Visits 2 and 3 were compared for statistical analysis in a within-subject comparison, meaning that test results after 4 weeks of FSP were compared with test results after 4 weeks of CIS in the same patient. The division into 2 groups was done to randomly distribute the sequence. The crossover design controls for effects of time (if

4 ENVELOPE VERSUS FINE STRUCTURE 1097 TABLE 1. Demographic data of all patients No. Sex Age at implantation (yr) Experience (yr) Age at follow-up (yr) Cause Implant 1 Female Progressive Pulsar 2 Female Progressive Pulsar 3 Male Progressive Pulsar 4 Female Meningitis Pulsar 5 Female Progressive Pulsar 6 Female Progressive Pulsar 7 Female Progressive Pulsar 8 Female Progressive Pulsar 9 Male Progressive Pulsar 10 Female progressive Pulsar 11 Female Progressive Pulsar 12 Male Progressive Pulsar 13 Female Progressive Pulsar 14 Female Progressive Pulsar 15 Male Progressive Pulsar 16 Male Progressive Pulsar 17 Female Progressive Pulsar 18 Male Progressive Pulsar 19 Female Progressive Pulsar 20 Female / Otosclerosis Sonata/Combi Male Progressive Pulsar 22 Male Progressive Pulsar 23 Male Progressive Pulsar 24 Female Progressive Pulsar 25 Male / Progressive Combi 40+/Sonata 26 Female Progressive Pulsar 27 Female Progressive Pulsar 28 Male Otosclerosis Pulsar 29 Male progressive Sonata 30 Female progressive Sonata 31 Male progressive Sonata Mean: results improve over time), which eliminates any learning effect (e.g., adaptation to test material, improvement of CI performance over the duration of the study). No between-group analysis was performed because it has less statistical power than the employed model accounting for the crossover design. Two subjects randomized to group B were excluded from the study after baseline visit tests. One patient had too little speech perception (30% correct score in a 2 digit numbers test), and the other patient refused to have any setting of the speech processor changed for study purposes. Their baseline scores and demographic details were discarded and not used Channel TABLE 2. Frequency bands Frequency range (Hz) 1 80Y Y Y Y Y Y1, ,374Y1, ,884Y2, ,600Y3, ,466Y4, ,799Y6, ,239Y8,466 This table shows corner frequencies of filter bands of the map used in this study from 70 to 8,500 Hz as displayed by the fitting software. The lowest frequency band is displayed starting at 80 Hz. for analysis. Altogether, 31 patientsv17 patients in group A and 14 patients in group BVcompleted the study. Their baseline scores (Visit 1) were only used as a covariate in the statistical model. Both groups displayed similar levels of baseline scores for HSM sentence tests: group A 32.2% (T6.2standarderrorof the mean [SEM]), group B 36.5% (T5.7 SEM), with p = 0.62 (unpaired t test). The primary end point of this study was the sentence recognition score in noise at 10 db signal-to-noise ratio (SNR). This test was chosen because, in a previous study (16), it was the test with the greatest observed difference between speech coding strategies and would therefore be most likely to show a significant difference. Secondary end points included monosyllabic word perception in quiet and melody recognition scores. Audiologic Testing Tests were taken from the standard stock of audiologic tests used at our department. All tests were presented from a CD in a sound-isolated room in free field. The Hochmair, Schulz and Moser (HSM) sentence test was conducted at 70 db in speech shaped noise at 10 db SNR (17). This test was designed to test speech intelligibility in noise with cochlear implants. It consists of lists with 20 sentences each and a total of 106 words per list. The lists are balanced to ensure equal difficulty with noise at 10dB SNR (17). According to protocol, a short list of 10 sentences was offered to patients to adapt to the intonation and speed of the speaker. Thereafter, one list of 20 sentences was tested at each visit. No repetition was permitted, and no feedback was provided.

5 1098 D. RISS ET AL. In addition, the Freiburger monosyllables test was presented at 70 db in quiet. This is an open set monosyllables test played from a CD. It consists of lists with 20 difficulty balanced monosyllabic words each. Three lists were presented to each listener. No repetition was permitted, and no feedback was provided. A music perception test was taken from the Mu.S.I.C. test battery (v1.0, MED-EL, Innsbruck Austria). The melody test was used to assess perception of melodies. Patients listen to 2 short melodies, either identical or different, which have to be correctly recognized by the listener. Therefore, the chance level of this experiment is 50%. Only melody pairs with similar rhythm and differing pitch contours were used. This ensured that no rhythmic cues were provided. These melodies were chosen because fine structure information is especially important for recognizing pitch contours (4). Twenty-eight patients were available for music test runs with both speech coding strategies. Any hearing aid in the contralateral ear was switched off. One patient with near-normal hearing in the contralateral ear was masked with a headphone introducing noise at 60 db SPL. The influence of possible residual low-frequency hearing was accounted for by using a within patient analysis, which ensures similar levels of possible low-frequency hearing for both speech coding strategies. Sample Size The HSM sentence test was the primary end point of this crossover study, and sample size calculation was based on this outcome measure. Estimated by an internal pilot study after completion of the first 16 patients, using the data at hand, a standard deviation of differences in the primary end point of 19.3 was assumed. Therefore, a required sample size of 32 patients was calculated to detect a clinically relevant difference of 10 percentage points with a statistical power of 80%, using the paired t test with a 0.05 two-sided significance level. As previously mentioned, 31 patients were randomized and completed the study. One patient with scheduled study visits quit the study before randomization for personal reasons. Statistical Analysis Results of the HSM sentence test, the Freiburger monosyllables test and the melody recognition test were analyzed using analysis of covariance (ANCOVA) models accounting for the crossover design of the study. Within these ANCOVA models, the speech coding strategy (FSP versus CIS) was tested, as well as potential time (first versus second period) and carryover effects. The time effect would reveal a possible learning effect (subjects achieve higher scores with each consecutive visit). This is adjusted for by including the time (Visit 2 or 3) at which tests were taken in the statistic model. A carryover effect tests an influence of the sequence whether the use of FSP before CIS influences the result achieved with CIS or vice versa. Furthermore, the baseline scoresvevaluated before randomizationv were considered as covariates in the ANCOVA models. All p values are results of 2-sided tests. A p value of less than 0.05 was considered to indicate statistical significance. RESULTS Primary End Point HSM sentence test results at70 db in 10 db SNR are shown in Figure 3. Mean results were 38.3% (T4.7 SEM) for CIS and 35.5% (T4.4 SEM) for FSP. Although the average levels were similar, there was a high variance among patients concerning the benefit obtained from one or the other speech coding strategy. This is presented together with baseline results in Figure 4. ANCOVA of the sentence test results showed no statistical significance for speech coding strategy (p = 0.36). No time or carryover effect was noted. Secondary End Points Results of monosyllabic word scores in quiet at 70 db show a similar distribution with pronounced differences between patients. On average, results were 43.2% (T3.7 SEM) for CIS and 39.9% (T3.9 SEM) for FSP. ANCOVA analysis revealed no statistical significance between coding strategies (p = 0.07). Again, there was no significant time or carryover effect. Results are displayed in Figure 5. Melody recognition experiments (Fig. 6) showed no differences between coding strategies in the ANCOVA FIG. 3. A, Results in percentages for the HSM sentence test in noise ( db SNR), visualized by box plots: results using FSP are on the left and CIS results on the right. Scores were measured after a 4-week interval of adaptation to the speech coding strategy in a crossover setting. Therefore, results with FSP include Visit 2 of group B and Visit 3 of group A, and results with CIS include Visit 3 of group B and Visit 2 of group A. B, Boxplots of results for Visits 2 and 3 separately. Patients using CIS (gray bar) at Visit 2 used FSP (white bar) at Visit 3 and vice versa. Statistical analysis by an ANCOVA model was not significant. The boundaries of the box represent the 25th and 75th percentile, respectively. The horizontal line inside the box represents the median. The whiskers indicate the last measurement within 1.5 interquartile ranges from the 25th or 75th percentiles, respectively. Circles represent values outside the box with a distance to the box of more than 1.5 times the box length.

6 ENVELOPE VERSUS FINE STRUCTURE 1099 Preference Patients were questioned about their preferences after completion of the trial. Approximately 64.5% chose to use an FSP map at the end of the study, 16.1% preferred the CIS map, and 19.4% considered both maps equal and asked to have both maps programmed on their processor to be able to switch according to listening situation. When patients were asked about their subjective experience, no specific listening situation was found, which would consistently favor one or the other strategy. DISCUSSION FIG. 4. Line charts of sentence scores showing individual results of each patient at each consecutive visit, including baseline scores at Visit 1. The left chart shows the results of group A, which was tested with CIS at Visit 2 and FSP at Visit 3, whereas the right chart shows group B, which used FSP first and CIS later. The values on the left of each chart (Visit 1) depict baseline values using FSP, which is the strategy all patients had been using before the study. Note the high variability of benefit with one or the other coding strategy between patients. Comparison of baseline scores between groups showed no statistical significance (p =0.62). analysis (p = 0.78). Mean values for CIS were 72.1% (T2.31 SEM) and 74.2% (T1.9 SEM) for FSP. Learning Effect Statistical analysis adjusted for effects of the time of testing. The ANCOVA analysis revealed no time effect in any of the tests (p ). Therefore, a learning effect influencing the results of this study can be excluded. The primary aim of this study was to compare the 2 speech coding strategies (FSP and CIS) with identical stimulation parameters by a crossover design and thereby answer the question whether the fine structure speech coding strategy per se is beneficial for speech perception in noise. The primary end point was sentence recognition in noise (Fig. 3). No statistically significant difference was observed, although a large number of patients were tested. Hence, both speech coding strategies deliver equal performance in noise. This end point was chosen because in 2 previous studies (15,16), conducted when patients switched to a new generation of speech processors (Opus), the greatest difference between the 2 coding strategies was measured at exactly this condition (HSM sentences in noise at 70 db with 10 db SNR). In these 2 previous studies, patients were tested at baseline with CIS and at follow-up visits exclusively with FSP. A statistically significant improvement of speech perception scores was noted with FSP. Similar results also were shown by an upgrade study in children by Lorens et al. (18). In addition to a possible learning effect, which might have augmented effects in those studies, the frequency spectrum offered in the new processor using FSP was extended down to 70 Hz, whereas the older processor (Tempo+) using CIS only stimulated down to 200 Hz. Those results considered together with the present study might indicate the importance of low-frequency information (down FIG. 5. A, Results in percentages for the monosyllables test in quiet, visualized by box plots: results using FSP are on the left and CIS results on the right. Scores were measured after a 4-week interval of adaptation to the speech coding strategy in a crossover setting. Therefore, results with FSP include Visit 2 of group B and Visit 3 of group A, and results with CIS include Visit 3 of group B and Visit 2 of group A. B, Boxplots of results for Visits 2 and 3 separately. Patients using CIS (gray bar) at Visit 2 used FSP (white bar) at Visit 3 and vice versa. Statistical analysis by an ANCOVA model was not significant. (For details on box plot parameters, see Fig. 3 legend).

7 1100 D. RISS ET AL. FIG. 6. A, Results in percentage correct for the melody recognition test, visualized by box plots: results using FSP are on the left and CIS results on the right. Presented melodies were identical in rhythm and differed only in pitch contour. Scores were measured after a 4-week interval of adaptation to the speech coding strategy in a crossover setting. B, Boxplots of results for Visits 2 and 3 separately. Patients using CIS (gray bar) at Visit 2 used FSP (white bar) at Visit 3 and vice versa. Statistical analysis by an ANCOVA model was not significant. (For details on box plot parameters, see Fig. 3 legend). to 70 Hz) for speech perception in cochlear implant recipients. In a previous study by our group comparing FSP and CIS at a reduced number of channels with tests performed immediately after fitting, we also observed no statistically significant difference at identical frequency ranges (19). A recent study by Vermeire et al. (20) reported a longterm effect of FSP in 22 patients with a minimum CI experience of 5.5 months. A significant improvement in speech perception in noise was observed 1 year after switchover from CIS with Tempo+ to FSP, which is consistent with previously published results (16). Outcomes were compared with a group of 10 patients, who had not been eligible to receive FSP because of a too low stimulation rate. These patients used CIS with an extended frequency spectrum. No statistically significant improvement was seen in this smaller group over 1 year of use. An acute comparison between extended spectrum CIS and limited spectrum CIS at the end of the study showed a trend toward better results with the extended frequency spectrum. This is consistent with the results of the present study. However, a comparison of extended spectrum FSP and limited spectrum CIS in the FSP group showed no difference between settings at a small sample size. One of the secondary end points of the present study, monosyllabic scores in quiet (Fig. 5), showed a slight trend in favor of CIS, albeit with no statistical significance. A further secondary end point was a melody recognition task, which showed no statistically significant difference between CIS and FSP. Only the subjective preference at the end of the study showed a majority of patients preferring the fine structure coding strategy. This though has to be interpreted with caution. At the beginning of the study, all patients were already used to their FSP map, and although they were subsequently fitted with a CIS map for 4 weeks, it is possible that they were still more inclined to favor FSP out of familiarity. What was interesting to note in this patient group was how they perceived the differences between coding strategies concerning subjective sound quality. Some patients noted a pronounced difference, most a minor change and others hardly noticed any difference at all. The subjective preference concerning speech intelligibility and music sound quality was recently investigated by Magnusson (21) in an upgrade study over a 2-year period. Blinded paired comparisons in 19 subjects after 2 years revealed no statistically significant difference in overall preference between speech coding strategies. However, marked differences between subjects were noted. In an additional test with identical frequency ranges, again, no overall difference was noted, but 6 of 16 subjects could reliably discriminate between FSP and CIS at an identical frequency range. Speech reception thresholds for sentences in noise, which were tested in 9 subjects, showed no statistically significant difference from baseline values at 2-year follow-up. In every prospective study of cochlear implant settings, the learning effect is an important issue, which has to be taken into account. Speech reception scores have been shown to improve over time, primarily during the first year of implant use (2). In addition, it is always possible that adaptation effects to the test materials used can occur. The present study s crossover design guarantees that one group of patients starts with one strategy, whereas the other group starts with the alternative. In this way, learning and adaptation effects can be eliminated, something which is not possible in other study designs where all patients undergo the same order of experimental settings (e.g., an A-B-A or A-B-A-B design). We acknowledge that experience with FSP (minimum of 1 year) before the start of the study might have favored FSP in this comparison. Because we observed no difference between the 2 coding strategies, we believe that this difference in experience before the start of this study did not influence our results. As previously mentioned, the results of this study viewed together with previous studies on the speech processor upgrade (15,16) might indicate that the low frequency range of 70 to 200 Hz improves speech perception in electric stimulation. In electric acoustic stimulation, the benefit of acoustic information in this frequency range is

8 ENVELOPE VERSUS FINE STRUCTURE 1101 well established (22,23). Nevertheless, the low-frequency acoustic information in these bimodal studies is of a completely different kind. Therefore, direct comparisons between acoustic versus electric coverage of low frequencies are not possible. Despite the suggested importance of the low-frequency region, the addition of fine structure information in the low frequencies in this study did not result in speech perception improvement when compared with the envelopebased strategy CIS. A recent study by Schatzer et al. (13) investigated speech perception in 12 Cantonese-speaking CI users using a fine structure coding strategy. Tests were performed immediately after fitting. The results showed no significant differences in the comparison between fine structure and CIS strategies. This is of interest because tonal languages particularly depend on the transmission of rapid variations in pitch for speech perception, and for this reason, additional fine structure should theoretically be beneficial to speech understanding. The importance of fine structure information has been demonstrated for the representation of speech sounds (4) and the representation of tone patterns in tonal languages (24). Many authors subsequently assumed that fine structure information is discarded in CIS. However, as pointed out by Wilson and Dorman (2) in an extensive review, there are several mechanisms for how fine structure could be presented in CIS. With a sufficient rate of stimulation, modulation waveforms may represent temporal fine structure up to 400 Hz, and because of overlapping filter bands, a channel balance cue might transmit fine structure in the higher frequencies. Wilson and Dorman conclude that it is not known to what extent envelope-based strategies transmit fine structure and whether other strategies can increase this. The present study did not test the transmission of fine structure but rather the patients performance in situations where this information is supposed to be needed. Speech recognition scores in noise were similar with the 2 strategies, and melody recognition scores also showed no differences when comparing envelope (CIS) and fine structure (FSP) strategy. The melodies used in these tests had identical rhythmical patterns, different only in pitch contour. For this reason, an improvement in the transmission of fine structure was expected to improve melody recognition. These results indicate that implant recipients might not be able to sufficiently use the fine structure cues offered by the FSP strategy. CONCLUSION This study offers a controlled comparison of the FSP strategy with the envelope-based strategy CIS at an identical frequency spectrum. Despite recent efforts to transmit fine structure, no advantage of one strategy over the other in either speech perception in quiet or noise or melody recognition could be observed. Therefore, the extended frequency spectrum in the low frequencies might explain a benefit of FSP observed in previous studies. REFERENCES 1. Wilson BS, Finley CC, Lawson DT, et al. Better speech recognition with cochlear implants. Nature 1991;352:236Y8. 2. Wilson BS, Dorman MF. Cochlear implants: a remarkable past and a brilliant future. Hear Res 2008;242:3Y Hilbert D. Grundzüge einer allgemeinen Theorie der linearen Integralgleichungen. Leipzig, Germany: Teubner, Smith ZM, Delgutte B, Oxenham AJ. Chimaeric sounds reveal dichotomies in auditory perception. Nature 2002;416:87Y Bonham BH, Litvak LM. Current focusing and steering: modeling, physiology, and psychophysics. Hear Res 2008;242:141Y Buechner A, Brendel M, Krueger B, et al. Current steering and results from novel speech coding strategies. Otol Neurotol 2008;29:203Y7. 7. Berenstein CK, Mens LH, Mulder JJ, et al. Current steering and current focusing in cochlear implants: comparison of monopolar, tripolar, and virtual channel electrode configurations. Ear Hear 2008;29:250Y Zeng FG, Rebscher S, Harrison WV, et al. Cochlear implants: system design, integration and evaluation. IEEE Rev Biomed Eng 2008; 1:115Y Zeng FG, Nie K, Stickney GS, et al. Speech recognition with amplitude and frequency modulations. Proc Natl Acad Sci U S A 2005;102:2293Y Nie K, Stickney G, Zeng FG. Encoding frequency modulation to improve cochlear implant performance in noise. IEEE Trans Biomed Eng 2005;52:64Y Throckmorton CS, Selin Kucukoglu M, Remus JJ, et al. Acoustic model investigation of a multiple carrier frequency algorithm for encoding fine frequency structure: implications for cochlear implants. Hear Res 2006;218:30Y Zierhofer CM, inventor; MED-EL Elektromedizinishce Geraete GmbH, assignee. Electrical nerve stimulation based on channel specific sampling sequences. US patent July 15, Schatzer R, Krenmayr A, Au DK, et al. Temporal fine structure in cochlear implants: preliminary speech perception results in Cantonesespeaking implant users. Acta Otolaryngol 2010;130:1031Y Brown CA, Bacon SP. Fundamental frequency and speech intelligibility in background noise. Hear Res 2010;266:52Y Arnoldner C, Riss D, Brunner M, et al. Speech and music perception with the new fine structure speech coding strategy: preliminary results. Acta Otolaryngol 2007;127:1298Y Riss D, Arnoldner C, Reiss S, et al. 1-year results using the Opus speech processor with the fine structure speech coding strategy. Acta Otolaryngol 2009;129:988Y Hochmair-Desoyer I, Schulz E, Moser L, et al. The HSM sentence test as a tool for evaluating the speech understanding in noise of cochlear implant users. Am J Otol 1997;18:S Lorens A, Zgoda M, Obrycka A, et al. Fine Structure Processing improves speech perception as well as objective and subjective benefits in pediatric MED-EL COMBI 40+ users. Int J Pediatr Otorhinolaryngol 2010;74:1372Y Riss D, Arnoldner C, Baumgartner WD, et al. A new fine structure speech coding strategy: speech perception at a reduced number of channels. Otol Neurotol 2008;29:784Y Vermeire K, Kleine Punte A, Van de Heyning P. Better speech recognition in noise with the fine structure processing coding strategy. ORL J Otorhinolaryngol Relat Spec 2010;72:305Y Magnusson L. Comparison of the fine structure processing (FSP) strategy and the CIS strategy used in the MED-EL cochlear implant system: Speech intelligibility and music sound quality. Int J Audiol 2011;50:279Y Zhang T, Spahr AJ, Dorman MF. Frequency overlap between electric and acoustic stimulation and speech-perception benefit in patients with combined electric and acoustic stimulation. Ear Hear 2010;31:195Y Buchner A, Schussler M, Battmer RD, et al. Impact of low-frequency hearing. Audiol Neurootol 2009;14:8Y Xu L, Pfingst BE. Relative importance of temporal envelope and fine structure in lexical-tone perception. J Acoust Soc Am 2003;114: 3024Y7.