| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Contact author: Sneha V. Bharadwaj, Callier Center, University of Texas at Dallas, 811 Synergy Park Boulevard, Richardson, TX 75080. E-mail: snehab{at}utdallas.edu.
 |
Abstract |
|---|
 Top Abstract References |
|---|
VOT (e.g., VOT /
/ – VOT /
/).
Method: Phonetic transcription and acoustic analyses were carried out on syllable–initial // and // produced by 10 children with cochlear implants. The DOVC was calculated as the difference between the shortest VOT value of a voiceless stop and the longest VOT value of a voiced stop across several productions of each.
Results: Phonetic transcription revealed that 4 of the 10 talkers demonstrated atypical voicing distinctions. Acoustic analyses indicated that the DOVC measure identified these same 4 talkers as producing atypical values, whereas mean VOT and VOT identified a different set of talkers as demonstrating values outside the normal ranges.
Conclusion: Preliminary findings suggest that the DOVC measure corresponded with perceptual data better than the other acoustic measures examined in the present study. Data indicate that the DOVC measure may provide perceptually relevant information concerning the production of voicing distinctions.
KEY WORDS: children, cochlear implants, voice onset time, production, acoustic
Voice onset time (VOT) is the duration between the release of a complete articulatory constriction and the onset of phonation (Lisker & Abramson, 1964). VOT is typically used to classify productions associated with voiced and voiceless stop consonants in syllable–initial position. Production of accurate voicing distinctions is assessed using both perceptual and acoustic measures. The perceptual measures include phonetic transcription and judgments from voicing identification experiments. Some of the acoustic measures include mean VOT duration, VOT range, VOT (mean voiceless VOT – mean voiced VOT), first formant frequency cutback, burst amplitude, and vowel duration (e.g., Forrest & Rockman, 1988; Jiang, Chen, & Alwan, 2006; Lisker & Abramson, 1964; Monsen, 1976; Ryalls & Larouche, 1992; Wambaugh, West, & Doyle, 1997).
Accurate production of voicing distinctions is impacted by hearing loss. For instance, through acoustic analysis, Monsen (1976) showed that more than half of the children with profound deafness between the ages of 11 and 16 years, who used hearing aids, did not produce voiceless versus voiced distinctions, whereas all their age-matched controls demonstrated accurate voicing distinctions. Further, Waldstein (1990) compared the production of voicing contrasts by individuals who had experienced hearing loss either during childhood (prelingually deafened) or as adults (postlingually deafened). Compared with individuals with normal hearing, the adults who were postlingually deafened demonstrated shortened VOT values for voiceless consonants. Similarly, the adults who were prelingually deafened demonstrated shortened VOT values along with other types of voicing errors, including a greater degree of prevoicing and overlapping VOT distributions. This finding suggests that the onset of deafness at younger ages is particularly detrimental to accurate production of voicing distinctions.
Cochlear implantation in children with severe–profound deafness has led to substantial short- and long-term improvement in the production of voicing distinctions (see Economou, Tartter, Chute, & Hellman, 1992; Higgins, McCleary, Carney, & Schulte, 2003; Lane, Wozniak, Matthies, Svirsky, & Perkell, 1995). In a study of 8- to 9-year-old users of cochlear implants, Uchanski and Geers (2003) showed mean VOT values, VOT range, and average VOT (VOT // – //) to be within normal limits in 84%–88% of talkers enrolled in oral communication programs compared with 62%–79% of talkers enrolled in total communication programs. Although these findings are encouraging, it appears that accurate production of voiced and voiceless consonants continues to be difficult for a subset of children with cochlear implants. Another recent study that examined the production of voicing distinctions in pediatric users of cochlear implants versus hearing aids showed comparable mean VOT values for both groups. However, the findings indicate that neither group differentiated voiced versus voiceless categories in their production as accurately as the control participants with normal hearing (see Horga & Liker, 2006). Considered together, both studies showed that a subgroup of children with cochlear implants did not demonstrate VOT values within the normal ranges. These results suggest a need for continued investigations of the production of voicing distinctions by children with hearing impairments.
One area pertaining to voicing production that needs further research is the identification of a good acoustic measure that corresponds to trained listeners' judgments of voicing errors. Studies of individuals with hearing impairments and phonological disorders have revealed that some of the commonly used acoustic measures, such as mean VOT duration, may not agree well with the perceptual judgments of voicing production. For example, Catts and Jensen (1983) showed that children judged by listeners with normal hearing to accurately distinguish between voiced and voiceless stops in their productions did not necessarily demonstrate VOT values comparable with their age-matched peers. In another study by Horga and Liker (2006), acoustic analysis did not reveal any group difference in VOT values in children using cochlear implants versus hearing aids. However, listeners judged voicing production in children with cochlear implants to be better than in children with hearing aids. Thus, there is an indication in the literature that some VOT-based acoustic measures may not correspond well with perceptual judgments of the production of voicing distinctions (see Forrest & Rockman, 1988, for a review). Therefore, one of the main objectives of the present study was to identify a VOT-based acoustic measure that corresponds with trained listeners' judgments of voicing distinctions.
In the present study, we evaluated the efficacy of an acoustic measure, called the discreteness of voicing category (DOVC), in identifying voicing errors in children with cochlear implants. The DOVC measure represents the difference between the shortest VOT value of a voiceless stop and the longest VOT value of a voiced stop across several productions of each. In a study of the development of the voicing contrast, Zlatin and Koenigsknecht (1976) showed a substantial overlap in VOT distributions of 2-year-old children (negative DOVC values), only a minimal overlap in the distributions of 6-year-old children, and no overlap (positive DOVC values) in the distributions of voiced and voiceless stop consonants produced by adults. Thus, the DOVC measure has been used to quantify the degree of overlap or separation in the VOT distributions (Zlatin & Koenigsknecht, 1976). Because the DOVC measure reflects the degree of overlap between the distributions of voiceless and voiced stops and takes into account the extreme values in the distributions, it was hypothesized that this measure would be closely related to trained listeners' judgments concerning the production of voicing distinctions. Because the measure is not widely used in speech production studies, the present study evaluated the DOVC measure along with conventional voicing measures to examine the extent to which several acoustic measures of voicing correspond with the perceptual measures of voicing distinctions.
A secondary objective of the present study was to examine VOT values of stop consonants produced by children fitted with cochlear implants in relation to normative data and to trained listeners' judgments of the voicing errors. Specifically, we examined average VOT values of plosives produced in response to the targets // and // by pediatric cochlear implant users. It was expected that children who demonstrated VOT values outside the normal ranges may not necessarily be identified by trained listeners for producing voicing errors.
To summarize, the present study explored the utility of the DOVC measure in identifying voicing errors in children with cochlear implants. In addition, this study examined the perceptual relevance of the DOVC measure and two commonly used VOT-based measures, namely, mean VOT and VOT. To that end, VOT measures of syllable–initial // and // were examined in the speech samples of 10 children fit with a cochlear implant using both phonetic transcription and acoustic analyses.
Method
Participants
Talkers. Participants included 10 children with prelingual deafness (S1–S10) who had cochlear implant experience of at least 4 years. Information concerning age, gender, age of implantation, length of implant use, and speech intelligibility scores (assessed as described in Tobey, Geers, Brenner, Altuna, & Gabbert, 2003) is reported in Table 1. Participants were monolingual speakers of American English and currently used the oral–aural mode of communication. Participants were paid for their participation in the study.
|
/ and //s. Two of the three transcribers were students enrolled in the communication disorders program at the University of Texas at Dallas. Both students were trained in phonetics. The third transcriber was a speech scientist who had several years of experience in teaching graduate and undergraduate phonetics courses.
Speech Materials and Procedure
Children were familiarized with the speech materials prior to completing the experimental task. Participants read the target words "tip" and "dip" embedded in the following carrier sentence: "It's a _______ again." The consonants // and // were chosen as stimuli because recent studies of children with cochlear implants have reported VOT values for these consonants (e.g., Uchanski & Geers, 2003). Participants read each target word from an index card, 12 times, in a random order, resulting in a total of 240 target productions (10 talkers x 2 syllables x 12 repetitions). For participant S2, only 11 repetitions of each target word were available for analysis because of poor recording quality.
Phonetic transcription. Phonetic transcription was used as a perceptual measure. The transcribers listened to the target words "tip" and "dip" in the context of a carrier sentence via headphones. Only the target sounds // and // were transcribed. The transcription was carried out in a two-step process. First, two graduate students performed a broad transcription of all occurrences of // and //s to identify the talkers who produced the target sounds inaccurately. The agreement between the transcribers was 100%. Second, the speech scientist performed narrow transcription of the syllable–initial // and //s produced by talkers that were identified in the first step as having made production errors. All transcribers were blind to the results of the acoustic data.
Acoustic analyses. VOT values were measured from both waveform and spectral displays via Brown Lab Interactive Speech System (BLISS) speech analysis software (Mertus, 2002). VOT was measured as the duration between the release of initial stop consonants and the onset of the first glottal pulse. For all productions of // with prevoicing, VOT was measured as the time lapse between the onset of prevoicing to the onset of the consonantal burst.
Mean VOT for // and // were determined for each talker by calculating the average VOT values across several repetitions of // and //. VOT for each talker was calculated by obtaining the difference between mean VOT for // and mean VOT for //. Finally, the DOVC measure was calculated for each talker by obtaining the difference between the lowest VOT value of // and the highest VOT value of // on the basis of 12 repetitions of // and //.
Mean VOT, VOT, and the DOVC measures were calculated from only those productions that represented correct manner of articulation. That is, on the basis of phonetic transcription data, only productions that represented correct manner of articulation ("good plosives") were selected for analysis. In other words, similar to that reported in Uchanski and Geers's (2003) study, if an intended plosive was produced with an incorrect place of articulation (e.g., /
/ or /
/), then that production was included for analysis. However, if the intended alveolar plosive was produced with a different manner (e.g., /
/), then that production was eliminated from the analysis. Table 2 shows the total number of tokens that were used for acoustic analysis, after the productions with incorrect manner were eliminated from the data set. Table 3 and Figure 1 (discussed later) reflect VOT data from tokens produced with correct manner. There were no noteworthy changes in the average VOT, VOT, or the DOVC values computed from all tokens versus tokens produced with only correct manner.
|
|
|
/ measured by the two judges was 1.3 ms (a measurement error of 1.45%). Similarly, the mean absolute difference in VOT for // measured by the two judges was 0.3 ms (a measurement error of 0.3%). A Pearson product–moment correlation coefficient was computed to examine the relationship between measurements from the two judges. The correlation was strong and highly significant (r = .99, p < .01), indicating high interjudge reliability.
The acoustic measures were subjected to several analyses. An analysis of variance was conducted to determine whether each talker produced a statistically significant difference in VOT values for voiced versus voiceless target sounds. In addition, mean VOT and VOT values for each talker were compared with normative data. Finally, mean VOT, DOVC, and VOT measures were examined in relation to phonetic transcription data.
Results and Discussion
Phonetic Transcription
The broad transcription carried out by the two graduate students identified 4 talkers (S4, S5, S6, and S8) as having produced voicing errors. After listening to the entire data set to confirm that the other talkers had produced the alveolar stop consonants correctly, the speech scientist performed narrow transcription of the target // and // productions of S4, S5, S6, and S8. Table 2 shows the number and types of errors that these talkers made. All 4 talkers made errors in // production, including aspiration, substitution of // with a fricative, and voicing errors. Only 2 talkers—S4 and S5—demonstrated errors in // production. These errors included excessive aspiration, substitution of // with a fricative, and voicing errors. Thus, phonetic transcription identified talkers S4, S5, S6, and S8 as producing errors in voicing, place, and manner of articulation.
It is interesting to note that talkers S4, S5, and S8, who demonstrated substantial errors (compared with S6, who produced only one error), also received poorer speech intelligibility scores than the rest of the talkers in the study (see Table 1). This finding perhaps indicates that the talkers who are able to produce voicing distinctions are also able to produce other features of speech correctly.
Acoustic Analyses
Mean VOT duration. Mean VOT values were analyzed in two ways. First, for each talker, a one-way analysis of variance was conducted using consonant as a factor to examine whether each talker produced a reliably different mean VOT for // versus //. On the basis of phonetic transcription, only tokens that were produced with correct manner were included in the analysis. The analysis revealed a significant main effect (p < .01) of consonant for all talkers except S5 and S8 (see Table 3). The results show that only 2 talkers—S5 and S8—did not reliably differentiate between voiced versus voiceless cognates. However, it should be noted that the transcribers identified 4 talkers—S4, S5, S6, and S8—as producing voicing errors. These results suggest that the approach of examining whether a talker produced reliably different VOT values for voiced versus voiceless stop consonants may not entirely agree with trained listeners' judgments of voicing errors.
Second, the mean VOT values for all talkers in the present study were compared with available normative data (Uchanski & Geers, 2003; Zlatin & Koenigsknecht, 1976). Table 4 shows distributional characteristics for VOT values of // and // tokens produced with correct manner by all participants. In addition, normative data for VOT values of alveolar stop consonants produced by 6- and 8- to 9-year-old children are shown in Table 4. As shown in Table 4, mean VOT values for talkers S1, S2, S4, S6, S7, and S8 (mean listening age of 5 years, 2 months) were compared with normative data provided by Zlatin and Koenigsknecht (1976) for 6-year-old children. Likewise, mean VOT values for // and // produced by talkers S3, S5, S9, and S10 (mean listening age of 8 years, 11 months) were compared with normative data provided by Uchanski and Geers (2003) for 8- to 9-year-old children. These comparisons revealed that the VOT values for // produced by talkers S3, S7, S8, and S9 differed from the normative data by more than one standard deviation. In addition, VOT values for // produced by talkers S4 and S8 differed from the normative data by more than one standard deviation.
|
/ and/or // outside normal ranges. However, phonetic transcription showed voicing production to be accurate in talkers S3, S7, and S9. As expected, findings from mean VOT data support those of Catts and Jensen (1983) and suggest that the talkers who are judged by transcribers as producing accurate voicing may not necessarily demonstrate VOT values similar to their age-matched controls. The findings suggest that listeners rely on several cues besides VOT duration for making judgments concerning accurate production of voicing distinctions.
VOT. Normative data for mean VOT reported by Zlatin and Koenigsknecht (1976) for children with normal hearing ranged approximately from 36 to 70 ms. Similarly, Uchanski and Geers (2003) reported a VOT range of 21–83 ms for 8- to 9-year-old children with normal hearing. Only 2 of the current talkers—S5 and S8—produced VOT values outside this range (16.1 ms and 11.63 ms, respectively). The rest of the talkers produced VOT within normal ranges. Thus, whereas transcribers identified S4, S5, S6, and S8 as producing voicing errors, only S5 and S8 produced delta VOT values outside the normal range. This finding suggests that VOT captures some but not all talkers who produce overlapping VOT distributions because it considers only mean VOT values and disregards outliers.
VOT distributions and DOVC. To plot the frequency distributions for each participant, the VOT values were grouped into 20-ms intervals. Figure 1 shows VOT distributions for participants S1–S10. The DOVC values are reported in the lower right corner of the distributions in Figure 1. Negative values reflect overlapping distributions, whereas positive values reflect nonoverlapping distributions. Talkers S4, S5, S6, and S8 demonstrated negative values, whereas the rest demonstrated positive DOVC values, reflecting well-separated distributions. Talker S6 showed a small negative value, suggesting minimal overlap. In fact, phonetic transcription also indicated only one instance of voicing error for this talker. The DOVC measure therefore identified the 4 talkers selected by the transcribers as producing voicing errors, suggesting that this measure corresponds well with the judgments provided by trained listeners.
One practical concern that could be raised with reference to the DOVC measure is whether it can be used in a clinical setting and whether the DOVC measure based on a smaller sample of voiced and voiceless consonants would also yield perceptually relevant information. To address this issue, the DOVC measure was recalculated for each talker using only the first five tokens of the words "tip" and "dip" (instead of the 12 tokens each). The results show negative DOVC values for the same 3 talkers (S4, S5, and S8) who had demonstrated substantial overlap between their // and // VOT distributions and who had been identified by transcribers as producing voicing errors. This finding should be confirmed in future studies to further evaluate the potential application of the DOVC measure in clinical settings.
In summary, the DOVC measure appears to be a perceptually relevant acoustic measure because it quantifies the separation of the voiced and voiceless distributions by taking into account the end points of the distributions. It is well established that more than one type of acoustic cue may lead to the same phonetic percept. Thus, a listener may not necessarily rely on just one cue. Nonetheless, it is striking that the DOVC measure captured that same set of talkers that the transcribers identified as producing voicing errors. Although these findings are important, they must be considered as preliminary and should be extended to a wider subset of consonants produced by a larger sample of talkers. In addition, it would be valuable to supplement the phonetic transcription data with an additional perception measure, such as a VOT identification experiment, to evaluate further the clinical/perceptual relevance of the DOVC measure.
Summary and Conclusions
The efficacy of the DOVC measure in identifying voicing errors was evaluated by examining consonants // and // produced by 10 pediatric cochlear implant users. Phonetic transcription of syllable–initial // and // was also carried out to assess how well various VOT-based acoustic measures (mean VOT, VOT, and the DOVC) correspond to trained listeners' judgments of voicing production accuracy. Phonetic transcription by three individuals with phonetic training revealed voicing errors in 4 of the 10 talkers. Of the acoustic measures assessed here, only the DOVC measure captured these same 4 talkers. Mean VOT and VOT measures identified a different subset of talkers who demonstrated VOT values outside normal ranges. Thus, the DOVC measure corresponded with the perceptual judgments by trained listeners better than any other acoustic measures examined in the present study. Although the sample size is limited, the present data indicate that the DOVC measure may yield perceptually relevant information concerning voicing errors. In addition, preliminary data suggest that the DOVC measure can be used to objectively evaluate VOT-related outcomes in clinical settings. Additional studies are necessary to extend the findings of the present study to a wider set of consonants produced by a large number of normally developing children and individuals with speech disorders.
|
Acknowledgments |
|---|
Received April 12, 2006
Revision received December 14, 2006
Accepted September 17, 2007
|
References |
|---|
|
TopAbstractReferences |
|---|
Catts, H. W., & Jensen, P. J. (1983). Speech timing of phonologically disordered children: Voicing contrast of initial and final stop consonants. Journal of Speech and Hearing Research, 26, 501–510.[Medline]
Economou, A., Tartter, V. C., Chute, P. M., & Hellman, S. A. (1992). Speech changes following reimplantation from a single-channel to a multichannel cochlear implant. The Journal of the Acoustical Society of America, 92, 1310–1323.[CrossRef][Medline]
Forrest, K., & Rockman, B. K. (1988). Acoustic and perceptual analysis of word-initial stop consonants in phonologically disordered children. Journal of Speech and Hearing Research, 31, 449–459.[Medline]
Higgins, M. B., McCleary, E. A., Carney, A. E., & Schulte, L. (2003). Longitudinal changes in children's speech and voice physiology after cochlear implantation. Ear and Hearing, 24, 48–70.[CrossRef][Medline]
Horga, D., & Liker, M. (2006). Voice and pronunciation of cochlear implant speakers. Clinical Linguistics & Phonetics, 20, 211–217.[CrossRef][Medline]
Jiang, J., Chen, M., & Alwan, A. (2006). On the perception of voicing in syllable–initial plosives in noise. The Journal of the Acoustical Society of America, 119, 1092–1105.[CrossRef][Medline]
Lane, H., Wozniak, J., Matthies, M., Svirsky, M., & Perkell, J. (1995). Phonemic resetting versus postural adjustments in the speech of cochlear implant users: An exploration of voice-onset time. The Journal of the Acoustical Society of America, 98, 3096–3106.[CrossRef][Medline]
Lisker, L., & Abramson, A. S. (1964). A cross language study of voicing in initial stops: Acoustical measurements. Word, 20, 384–422.
Mertus, J. (2002). Brown Lab Interactive Speech System (BLISS) [Computer software]. Providence, RI: Brown University.
Monsen, R. B. (1976). The production of English stop consonants in the speech of deaf children. Journal of Phonetics, 4, 29–41.
Ryalls, J., & Larouche, A. (1992). Acoustic integrity of speech production in children with moderate and severe hearing impairment. Journal of Speech and Hearing Research, 35, 88–95.[Medline]
Tobey, E. A., Geers, A. E., Brenner, C., Altuna, D., & Gabbert, G. (2003). Factors associated with development of speech production skills in children implanted before age five. Ear and Hearing, 24, 36S–45S.[CrossRef][Medline]
Uchanski, R. M., & Geers, A. E. (2003). Acoustic characteristics of the speech of young cochlear implant users: A comparison with normal-hearing age-mates. Ear and Hearing, 24, 90S–105S.[CrossRef][Medline]
Waldstein, R. S. (1990). Effects of postlingual deafness on speech production: Implications for the role of auditory feedback. The Journal of the Acoustical Society of America, 88, 2099–2114.[CrossRef][Medline]
Wambaugh, J. L., West, J. E., & Doyle, P. J. (1997). A VOT analysis of apraxic/aphasic voicing errors. Aphasiology, 11, 521–532.[CrossRef]
Zlatin, M. A., & Koenigsknecht, R. A. (1976). Development of the voicing contrast: A comparison of voice onset time in stop perception and production. Journal of Speech and Hearing Research, 19, 93–111.[Medline]
CiteULike 
Connotea 
Del.icio.us 
Digg 
Facebook 
Reddit 
Technorati 
Twitter What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| All ASHA Journals | AJA | AJSLP | JSLHR | LSHSS |
