Efficacy of the Discreteness of Voicing Category (DOVC) Measure by zvf19883


									                                                    Efficacy of the Discreteness of Voicing
                                                    Category (DOVC) Measure for
                                                    Characterizing Voicing Errors in Children
                                                    With Cochlear Implants: A Report

                    Sneha V. Bharadwaj
                     Amanda G. Graves
                                                    Purpose: This investigation explored the utility of an acoustic measure, called the
Callier Center, University of Texas at Dallas
                                                    discreteness of voicing category ( DOVC), in identifying voicing errors in stop
                                                    consonants produced by children with cochlear implants. Another objective was to
                                                    examine the perceptual relevance of the DOVC measure and 2 commonly used
                                                    voice onset time ( VOT )-based measures, namely, mean VOT and DVOT (e.g., VOT
                                                    /t/ – VOT /d/).
                                                    Method: Phonetic transcription and acoustic analyses were carried out on syllable–
                                                    initial /t/ and /d/ 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 DVOT
                                                    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

                                                     V       oice onset time (VOT) is the duration between the release of a com-
                                                             plete 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 pho-
                                                    netic transcription and judgments from voicing identification experiments.
                                                    Some of the acoustic measures include mean VOT duration, VOT range,
                                                    DVOT (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

      Journal of Speech, Language, and Hearing Research • Vol. 51 • 629–635 • June 2008 • D American Speech-Language-Hearing Association   629
                                         Downloaded from jslhr.asha.org on August 13, 2010
accurate voicing distinctions. Further, Waldstein (1990)                  and Liker (2006), acoustic analysis did not reveal any
compared the production of voicing contrasts by indi-                     group difference in VOT values in children using cochlear
viduals who had experienced hearing loss either during                    implants versus hearing aids. However, listeners judged
childhood (prelingually deafened) or as adults (post-                     voicing production in children with cochlear implants to
lingually deafened). Compared with individuals with nor-                  be better than in children with hearing aids. Thus, there
mal hearing, the adults who were postlingually deafened                   is an indication in the literature that some VOT-based
demonstrated shortened VOT values for voiceless con-                      acoustic measures may not correspond well with percep-
sonants. Similarly, the adults who were prelingually                      tual judgments of the production of voicing distinctions
deafened demonstrated shortened VOT values along                          (see Forrest & Rockman, 1988, for a review). Therefore,
with other types of voicing errors, including a greater de-               one of the main objectives of the present study was to
gree of prevoicing and overlapping VOT distributions.                     identify a VOT-based acoustic measure that corresponds
This finding suggests that the onset of deafness at youn-                 with trained listeners’ judgments of voicing distinctions.
ger ages is particularly detrimental to accurate produc-                       In the present study, we evaluated the efficacy of an
tion of voicing distinctions.                                             acoustic measure, called the discreteness of voicing cat-
     Cochlear implantation in children with severe–                       egory ( DOVC), in identifying voicing errors in children
profound deafness has led to substantial short- and                       with cochlear implants. The DOVC measure represents
long-term improvement in the production of voicing                        the difference between the shortest VOT value of a voice-
distinctions (see Economou, Tartter, Chute, & Hellman,                    less stop and the longest VOT value of a voiced stop across
1992; Higgins, McCleary, Carney, & Schulte, 2003;                         several productions of each. In a study of the development
Lane, Wozniak, Matthies, Svirsky, & Perkell, 1995). In                    of the voicing contrast, Zlatin and Koenigsknecht (1976)
a study of 8- to 9-year-old users of cochlear implants,                   showed a substantial overlap in VOT distributions of
Uchanski and Geers (2003) showed mean VOT values,                         2-year-old children (negative DOVC values), only a min-
VOT range, and average DVOT (VOT /t/ – /d/) to be within                  imal overlap in the distributions of 6-year-old children,
normal limits in 84%–88% of talkers enrolled in oral com-                 and no overlap (positive DOVC values) in the distribu-
munication programs compared with 62%–79% of talkers                      tions of voiced and voiceless stop consonants produced
enrolled in total communication programs. Although                        by adults. Thus, the DOVC measure has been used to
these findings are encouraging, it appears that accurate                  quantify the degree of overlap or separation in the VOT
production of voiced and voiceless consonants continues                   distributions (Zlatin & Koenigsknecht, 1976). Because
to be difficult for a subset of children with cochlear im-                the DOVC measure reflects the degree of overlap be-
plants. Another recent study that examined the produc-                    tween the distributions of voiceless and voiced stops and
tion of voicing distinctions in pediatric users of cochlear               takes into account the extreme values in the distributions,
implants versus hearing aids showed comparable mean                       it was hypothesized that this measure would be closely
VOT values for both groups. However, the findings in-                     related to trained listeners’ judgments concerning the
dicate that neither group differentiated voiced versus                    production of voicing distinctions. Because the measure
voiceless categories in their production as accurately as                 is not widely used in speech production studies, the
the control participants with normal hearing (see Horga                   present study evaluated the DOVC measure along with
& Liker, 2006). Considered together, both studies showed                  conventional voicing measures to examine the extent to
that a subgroup of children with cochlear implants did                    which several acoustic measures of voicing correspond
not demonstrate VOT values within the normal ranges.                      with the perceptual measures of voicing distinctions.
These results suggest a need for continued investiga-                          A secondary objective of the present study was to
tions of the production of voicing distinctions by children               examine VOT values of stop consonants produced by
with hearing impairments.                                                 children fitted with cochlear implants in relation to nor-
     One area pertaining to voicing production that needs                 mative data and to trained listeners’ judgments of the
further research is the identification of a good acoustic                 voicing errors. Specifically, we examined average VOT
measure that corresponds to trained listeners’ judgments                  values of plosives produced in response to the targets /t/
of voicing errors. Studies of individuals with hearing im-                and /d/ by pediatric cochlear implant users. It was ex-
pairments and phonological disorders have revealed that                   pected that children who demonstrated VOT values out-
some of the commonly used acoustic measures, such as                      side the normal ranges may not necessarily be identified
mean VOT duration, may not agree well with the per-                       by trained listeners for producing voicing errors.
ceptual judgments of voicing production. For example,                          To summarize, the present study explored the util-
Catts and Jensen (1983) showed that children judged by                    ity of the DOVC measure in identifying voicing errors in
listeners with normal hearing to accurately distinguish                   children with cochlear implants. In addition, this study
between voiced and voiceless stops in their productions                   examined the perceptual relevance of the DOVC measure
did not necessarily demonstrate VOT values comparable                     and two commonly used VOT-based measures, namely,
with their age-matched peers. In another study by Horga                   mean VOT and DVOT. To that end, VOT measures of

630   Journal of Speech, Language, and Hearing Research • Vol. 51 • 629–635 • June 2008

                                      Downloaded from jslhr.asha.org on August 13, 2010
syllable–initial /t/ and /d/ were examined in the speech                     to the target words “tip” and “dip” in the context of a
samples of 10 children fit with a cochlear implant using                     carrier sentence via headphones. Only the target sounds
both phonetic transcription and acoustic analyses.                           /t/ and /d/ were transcribed. The transcription was car-
                                                                             ried out in a two-step process. First, two graduate stu-
                                                                             dents performed a broad transcription of all occurrences
                                                                             of /t/ and /d/s to identify the talkers who produced the
Method                                                                       target sounds inaccurately. The agreement between the
Participants                                                                 transcribers was 100%. Second, the speech scientist per-
                                                                             formed narrow transcription of the syllable–initial /t/
     Talkers. Participants included 10 children with pre-
                                                                             and /d/s produced by talkers that were identified in the
lingual deafness (S1–S10) who had cochlear implant
                                                                             first step as having made production errors. All tran-
experience of at least 4 years. Information concerning
                                                                             scribers were blind to the results of the acoustic data.
age, gender, age of implantation, length of implant use,
and speech intelligibility scores (assessed as described in                      Acoustic analyses. VOT values were measured from
Tobey, Geers, Brenner, Altuna, & Gabbert, 2003) is                           both waveform and spectral displays via Brown Lab Inter-
reported in Table 1. Participants were monolingual speak-                    active Speech System ( BLISS) speech analysis software
ers of American English and currently used the oral–aural                    (Mertus, 2002). VOT was measured as the duration
mode of communication. Participants were paid for their                      between the release of initial stop consonants and the
participation in the study.                                                  onset of the first glottal pulse. For all productions of /d/
                                                                             with prevoicing, VOT was measured as the time lapse
    Phonetic transcribers. Three volunteers performed
                                                                             between the onset of prevoicing to the onset of the con-
phonetic transcription of the syllable–initial /t/ and /d/s.
                                                                             sonantal burst.
Two of the three transcribers were students enrolled in
the communication disorders program at the University                             Mean VOT for /t/ and /d/ were determined for each
of Texas at Dallas. Both students were trained in pho-                       talker by calculating the average VOT values across
netics. The third transcriber was a speech scientist who                     several repetitions of /t/ and /d/. DVOT for each talker
had several years of experience in teaching graduate and                     was calculated by obtaining the difference between mean
undergraduate phonetics courses.                                             VOT for /t/ and mean VOT for /d/. Finally, the DOVC
                                                                             measure was calculated for each talker by obtaining the
                                                                             difference between the lowest VOT value of /t/ and the
Speech Materials and Procedure                                               highest VOT value of /d/ on the basis of 12 repetitions of
     Children were familiarized with the speech mate-                        /t/ and /d/.
rials prior to completing the experimental task. Partici-                         Mean VOT, DVOT, and the DOVC measures were
pants read the target words “tip” and “dip” embedded in                      calculated from only those productions that represented
the following carrier sentence: “It’s a _______ again.”                      correct manner of articulation. That is, on the basis of
The consonants /t/ and /d/ were chosen as stimuli be-                        phonetic transcription data, only productions that rep-
cause recent studies of children with cochlear implants                      resented correct manner of articulation (“good plosives”)
have reported VOT values for these consonants (e.g.,                         were selected for analysis. In other words, similar to that
Uchanski & Geers, 2003). Participants read each target                       reported in Uchanski and Geers’s (2003) study, if an
word from an index card, 12 times, in a random order, re-                    intended plosive was produced with an incorrect place of
sulting in a total of 240 target productions (10 talkers ×                   articulation (e.g., /p / or /g /), then that production was in-
2 syllables × 12 repetitions). For participant S2, only                      cluded for analysis. However, if the intended alveolar
11 repetitions of each target word were available for anal-                  plosive was produced with a different manner (e.g., /s/),
ysis because of poor recording quality.                                      then that production was eliminated from the analysis.
    Phonetic transcription. Phonetic transcription was                       Table 2 shows the total number of tokens that were used
used as a perceptual measure. The transcribers listened                      for acoustic analysis, after the productions with incorrect

Table 1. Demographic information for 10 pediatric users (S1–S10) of cochlear implants.

                  Variable                          S1       S2      S3         S4      S5        S6        S7         S8         S9        S10

Gender                                            Female   Female   Female     Male    Male    Female     Female     Female    Female       Male
Age (years;months)                                  7;2      7;9     10;0      10;2    10;4     10;6       12;0       13;8     13;10        15;3
Implanted age (years;months)                        2;1      3;9      2;0       6;0     2;7      5;0        5;11       7;7       5;0         3;0
Length of cochlear implant use (years;months)       5;1      4;0      8;0       4;2     7;7      5;6        6;1        6;1       8;10       11;5
Speech intelligibility (%)                          95       92       99        36      84       94         86         47        89          84

                                                                                         Bharadwaj & Graves: Efficacy of the DOVC Measure    631

                                                Downloaded from jslhr.asha.org on August 13, 2010
Table 2. Production errors in response to targets /t/ and /d/ by              for each talker were compared with normative data.
talkers S4, S5, S6, and S8.                                                   Finally, mean VOT, DOVC, and DVOT measures were
                                                                              examined in relation to phonetic transcription data.
                                                            No. of tokens
                                                             chosen for
Talker Target No. of errors        Substitutions          acoustic analysis
                                                                              Results and Discussion
S4         /t/       1/12                [h]                         11       Phonetic Transcription
           /d/       8/12                [t]                         12
                                                                                   The broad transcription carried out by the two grad-
S5         /t/      12/12           [d], [v], or [z]                 4
                                                                              uate students identified 4 talkers (S4, S5, S6, and S8) as
           /d/      10/12       [d ], [dh], [t], or [h]
                                 w                                   4
                                                                              having produced voicing errors. After listening to the en-
S6         /t/       0/12                                            12       tire data set to confirm that the other talkers had pro-
           /d/       1/12                [t]                         12       duced the alveolar stop consonants correctly, the speech
S8         /t/       0/12                                            12       scientist performed narrow transcription of the target /t/
           /d/      11/12                 [t]                        12       and /d/ 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 /d/ production, including as-
manner were eliminated from the data set. Table 3 and                         piration, substitution of /d/ with a fricative, and voicing
Figure 1 (discussed later) reflect VOT data from tokens                       errors. Only 2 talkers—S4 and S5—demonstrated er-
produced with correct manner. There were no noteworthy                        rors in /t/ production. These errors included excessive
changes in the average VOT, DVOT, or the DOVC values                          aspiration, substitution of /t/ with a fricative, and voic-
computed from all tokens versus tokens produced with                          ing errors. Thus, phonetic transcription identified
only correct manner.                                                          talkers S4, S5, S6, and S8 as producing errors in voicing,
     To establish interjudge reliability, 10% of the data                     place, and manner of articulation.
were reanalyzed by a second judge. The mean absolute                               It is interesting to note that talkers S4, S5, and S8,
difference in VOT for /t/ measured by the two judges was                      who demonstrated substantial errors (compared with
1.3 ms (a measurement error of 1.45%). Similarly, the                         S6, who produced only one error), also received poorer
mean absolute difference in VOT for /d/ measured by the                       speech intelligibility scores than the rest of the talkers
two judges was 0.3 ms (a measurement error of 0.3%). A                        in the study (see Table 1). This finding perhaps indicates
Pearson product–moment correlation coefficient was com-                       that the talkers who are able to produce voicing dis-
puted to examine the relationship between measurements                        tinctions are also able to produce other features of speech
from the two judges. The correlation was strong and highly                    correctly.
significant (r = .99, p < .01), indicating high interjudge
                                                                              Acoustic Analyses
    The acoustic measures were subjected to several
analyses. An analysis of variance was conducted to deter-                          Mean VOT duration. Mean VOT values were an-
mine whether each talker produced a statistically signif-                     alyzed in two ways. First, for each talker, a one-way
icant difference in VOT values for voiced versus voiceless                    analysis of variance was conducted using consonant as a
target sounds. In addition, mean VOT and DVOT values                          factor to examine whether each talker produced a reliably
                                                                              different mean VOT for /t/ versus /d/. On the basis of
                                                                              phonetic transcription, only tokens that were produced
     Table 3. Analysis of variance results for talkers who showed a           with correct manner were included in the analysis. The
     significant difference (p < .01) between mean voice onset time           analysis revealed a significant main effect ( p < .01) of
     values of stop consonants produced with correct manner in
                                                                              consonant for all talkers except S5 and S8 (see Table 3).
     response to targets /t/ and /d/.
                                                                              The results show that only 2 talkers—S5 and S8—did
                                                                              not reliably differentiate between voiced versus voice-
     Talker                                               F ratios
                                                                              less cognates. However, it should be noted that the tran-
     S1                                           F (1, 11) =   256.3         scribers identified 4 talkers—S4, S5, S6, and S8—as
     S2                                           F (1, 10) =   49.1          producing voicing errors. These results suggest that the
     S3                                           F (1, 11) =   223.2         approach of examining whether a talker produced re-
     S4                                           F (1, 10) =   19.4          liably different VOT values for voiced versus voiceless
     S6                                           F (1, 11) =   118.5         stop consonants may not entirely agree with trained
     S7                                           F (1, 11) =   299.6         listeners’ judgments of voicing errors.
     S9                                           F (1, 11) =   1321.9
     S10                                          F (1, 11) =   275.8             Second, the mean VOT values for all talkers in the
                                                                              present study were compared with available normative

632     Journal of Speech, Language, and Hearing Research • Vol. 51 • 629–635 • June 2008

                                            Downloaded from jslhr.asha.org on August 13, 2010
Figure 1. Voice onset time (VOT) distributions for targets /t/ and /d/ produced by participants S1–S10. The distributions represent VOT values
grouped into 20-ms bins for tokens produced with only correct manner. Values associated with the discreteness of voicing category are shown on the
lower right corner of the VOT distributions. An asterisk indicates a significant difference (p < .01) between mean VOT values for /t/ versus /d/.

data (Uchanski & Geers, 2003; Zlatin & Koenigsknecht,                           In all, talkers S3, S4, S7, S8, and S9 demonstrated
1976). Table 4 shows distributional characteristics for                     mean VOT values for /t/ and /or /d / outside normal
VOT values of /t/ and /d/ tokens produced with correct                      ranges. However, phonetic transcription showed voicing
manner by all participants. In addition, normative data                     production to be accurate in talkers S3, S7, and S9. As
for VOT values of alveolar stop consonants produced by                      expected, findings from mean VOT data support those of
6- and 8- to 9-year-old children are shown in Table 4. As                   Catts and Jensen (1983) and suggest that the talkers
shown in Table 4, mean VOT values for talkers S1, S2, S4,                   who are judged by transcribers as producing accurate
S6, S7, and S8 (mean listening age of 5 years, 2 months)                    voicing may not necessarily demonstrate VOT values
were compared with normative data provided by Zlatin                        similar to their age-matched controls. The findings
and Koenigsknecht (1976) for 6-year-old children. Like-                     suggest that listeners rely on several cues besides VOT
wise, mean VOT values for /t/ and /d/ produced by talkers                   duration for making judgments concerning accurate
S3, S5, S9, and S10 (mean listening age of 8 years,                         production of voicing distinctions.
11 months) were compared with normative data provided                           DVOT. Normative data for mean DVOT reported by
by Uchanski and Geers (2003) for 8- to 9-year-old chil-                     Zlatin and Koenigsknecht (1976) for children with nor-
dren. These comparisons revealed that the VOT values                        mal hearing ranged approximately from 36 to 70 ms.
for /t/ produced by talkers S3, S7, S8, and S9 differed from                Similarly, Uchanski and Geers (2003) reported a DVOT
the normative data by more than one standard deviation.                     range of 21–83 ms for 8- to 9-year-old children with nor-
In addition, VOT values for /d/ produced by talkers S4 and                  mal hearing. Only 2 of the current talkers—S5 and S8—
S8 differed from the normative data by more than one                        produced DVOT values outside this range (16.1 ms and
standard deviation.                                                         11.63 ms, respectively). The rest of the talkers produced

                                                                                         Bharadwaj & Graves: Efficacy of the DOVC Measure    633

                                            Downloaded from jslhr.asha.org on August 13, 2010
Table 4. Means (M), standard deviations (SD), voice onset time (VOT) range, and the discreteness of voicing category (DOVC) values for tokens
produced with correct manner in response to the targets /t/ and /d/ by participants S1–S10.

                                     /t/                                              /d/

Participant       M (ms)           SD            Range (ms)        M (ms)           SD         Range (ms)          DVOT (ms)          DOVC

S1                107.6           18.55            66.06            31.5           5.15           19                  76.0            35.9
S2                 80.82          27.04            71.04            17.72          9.42           30                  63.1            13.1
S4                 86.80          26.62            88.48            50.02         33.54          100.9                36.77          –58.8
S6                 82.54          15.06            50.41            26.69          9.05           39.80               55.8            –1.2
S7                115.3           21.72            82.84            16.98          3.99           14.67               98.3            73.9
S8                114.60          15.45            55.57           102.97         30.19          117.55               11.63          –36.89
ZK                 68.4           23.2            120               12.1          20.1           120                  56.3            –1.00
S3                109.9           23.41            85.53             7.82          1.99            5.44              102.1            55.0
S5                 45.22          16.89            40.61            29.09         15.58           35.87               16.13          –27.5
S9                106.2           11.88            39.86           –36.11          7.51           26.03              142.3           102.6
S10                69.01          11.36            38.43            16.19          3.63           11.45               52.8            20.1
UG                 72             15               61               17             5              24                  55              15

Note. ZK = normative data from Zlatin and Koenigsknecht’s (1976) study for 6-year-old children; UG = normative data from Uchanski and Geers’s
(2003) study for 8- to 9-year-old children.

DVOT within normal ranges. Thus, whereas transcribers                       for the same 3 talkers (S4, S5, and S8) who had dem-
identified S4, S5, S6, and S8 as producing voicing errors,                  onstrated substantial overlap between their /t/ and /d/
only S5 and S8 produced delta VOT values outside the                        VOT distributions and who had been identified by tran-
normal range. This finding suggests that DVOT captures                      scribers as producing voicing errors. This finding should
some but not all talkers who produce overlapping VOT                        be confirmed in future studies to further evaluate the
distributions because it considers only mean VOT values                     potential application of the DOVC measure in clinical
and disregards outliers.                                                    settings.
     VOT distributions and DOVC. To plot the frequency                            In summary, the DOVC measure appears to be a per-
distributions for each participant, the VOT values were                     ceptually relevant acoustic measure because it quantifies
grouped into 20-ms intervals. Figure 1 shows VOT dis-                       the separation of the voiced and voiceless distributions by
tributions for participants S1–S10. The DOVC values                         taking into account the end points of the distributions.
are reported in the lower right corner of the distributions                 It is well established that more than one type of acoustic
in Figure 1. Negative values reflect overlapping distri-                    cue may lead to the same phonetic percept. Thus, a lis-
butions, whereas positive values reflect nonoverlapping                     tener may not necessarily rely on just one cue. None-
distributions. Talkers S4, S5, S6, and S8 demonstrated                      theless, it is striking that the DOVC measure captured
negative values, whereas the rest demonstrated positive                     that same set of talkers that the transcribers identified
DOVC values, reflecting well-separated distributions.                       as producing voicing errors. Although these findings are
Talker S6 showed a small negative value, suggesting                         important, they must be considered as preliminary and
minimal overlap. In fact, phonetic transcription also in-                   should be extended to a wider subset of consonants pro-
dicated only one instance of voicing error for this talker.                 duced by a larger sample of talkers. In addition, it would
The DOVC measure therefore identified the 4 talkers                         be valuable to supplement the phonetic transcription
selected by the transcribers as producing voicing errors,                   data with an additional perception measure, such as a
suggesting that this measure corresponds well with the                      VOT identification experiment, to evaluate further the
judgments provided by trained listeners.                                    clinical/perceptual relevance of the DOVC measure.
    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
                                                                            Summary and Conclusions
based on a smaller sample of voiced and voiceless con-                           The efficacy of the DOVC measure in identifying
sonants would also yield perceptually relevant informa-                     voicing errors was evaluated by examining consonants
tion. To address this issue, the DOVC measure was                           /t/ and /d/ produced by 10 pediatric cochlear implant
recalculated for each talker using only the first five                      users. Phonetic transcription of syllable–initial /t/ and
tokens of the words “tip” and “dip” (instead of the                         /d/ was also carried out to assess how well various VOT-
12 tokens each). The results show negative DOVC values                      based acoustic measures (mean VOT, DVOT, and the

634     Journal of Speech, Language, and Hearing Research • Vol. 51 • 629–635 • June 2008

                                           Downloaded from jslhr.asha.org on August 13, 2010
DOVC) correspond to trained listeners’ judgments of voic-         Horga, D., & Liker, M. (2006). Voice and pronunciation of
ing production accuracy. Phonetic transcription by three           cochlear implant speakers. Clinical Linguistics & Phonetics,
                                                                   20, 211–217.
individuals with phonetic training revealed voicing er-
rors in 4 of the 10 talkers. Of the acoustic measures             Jiang, J., Chen, M., & Alwan, A. (2006). On the perception of
                                                                    voicing in syllable–initial plosives in noise. The Journal of
assessed here, only the DOVC measure captured these                 the Acoustical Society of America, 119, 1092–1105.
same 4 talkers. Mean VOT and DVOT measures iden-
                                                                  Lane, H., Wozniak, J., Matthies, M., Svirsky, M., &
tified a different subset of talkers who demonstrated              Perkell, J. (1995). Phonemic resetting versus postural
VOT values outside normal ranges. Thus, the DOVC                   adjustments in the speech of cochlear implant users: An
measure corresponded with the perceptual judgments                 exploration of voice-onset time. The Journal of the Acous-
by trained listeners better than any other acoustic mea-           tical Society of America, 98, 3096–3106.
sures examined in the present study. Although the sam-            Lisker, L., & Abramson, A. S. (1964). A cross language study
ple size is limited, the present data indicate that the             of voicing in initial stops: Acoustical measurements. Word,
DOVC measure may yield perceptually relevant infor-                 20, 384–422.
mation concerning voicing errors. In addition, prelimi-           Mertus, J. (2002). Brown Lab Interactive Speech System
nary data suggest that the DOVC measure can be used                ( BLISS) [Computer software]. Providence, RI: Brown
to objectively evaluate VOT-related outcomes in clinical
settings. Additional studies are necessary to extend the          Monsen, R. B. (1976). The production of English stop consonants
                                                                   in the speech of deaf children. Journal of Phonetics, 4, 29–41.
findings of the present study to a wider set of consonants
produced by a large number of normally developing chil-           Ryalls, J., & Larouche, A. (1992). Acoustic integrity of
                                                                   speech production in children with moderate and severe
dren and individuals with speech disorders.                        hearing impairment. Journal of Speech and Hearing
                                                                   Research, 35, 88–95.

Acknowledgments                                                   Tobey, E. A., Geers, A. E., Brenner, C., Altuna, D., &
                                                                   Gabbert, G. (2003). Factors associated with development of
     This work was supported by National Institute on Deaf-        speech production skills in children implanted before age
ness and Other Communication Disorders Grant R03DC007052           five. Ear and Hearing, 24, 36S–45S.
awarded to the first author. The authors would like to thank      Uchanski, R. M., & Geers, A. E. (2003). Acoustic character-
participants and their families for their contributions. Thanks    istics of the speech of young cochlear implant users: A com-
to Deborah Rekart and Anupama Jayaraman for their help in          parison with normal-hearing age-mates. Ear and Hearing,
phonetic transcription work. Thanks also to William F. Katz        24, 90S–105S.
and Raksha Anand for their helpful comments on an earlier         Waldstein, R. S. (1990). Effects of postlingual deafness on
version of this article.                                           speech production: Implications for the role of auditory
                                                                   feedback. The Journal of the Acoustical Society of America,
                                                                   88, 2099–2114.
References                                                        Wambaugh, J. L., West, J. E., & Doyle, P. J. (1997). A VOT
                                                                   analysis of apraxic/aphasic voicing errors. Aphasiology, 11,
Catts, H. W., & Jensen, P. J. (1983). Speech timing of             521–532.
 phonologically disordered children: Voicing contrast of
 initial and final stop consonants. Journal of Speech and         Zlatin, M. A., & Koenigsknecht, R. A. (1976). Development
 Hearing Research, 26, 501–510.                                     of the voicing contrast: A comparison of voice onset time
                                                                    in stop perception and production. Journal of Speech and
Economou, A., Tartter, V. C., Chute, P. M., & Hellman, S. A.        Hearing Research, 19, 93–111.
 (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.     Received April 12, 2006
Forrest, K., & Rockman, B. K. (1988). Acoustic and percep-        Revision received December 14, 2006
 tual analysis of word-initial stop consonants in phonologi-
                                                                  Accepted September 17, 2007
 cally disordered children. Journal of Speech and Hearing
 Research, 31, 449–459.                                           DOI: 10.1044/1092-4388(2008/045)
Higgins, M. B., McCleary, E. A., Carney, A. E., & Schulte,        Contact author: Sneha V. Bharadwaj, Callier Center,
 L. (2003). Longitudinal changes in children’s speech and           University of Texas at Dallas, 811 Synergy Park Boulevard,
 voice physiology after cochlear implantation. Ear and              Richardson, TX 75080. E-mail: snehab@utdallas.edu.
 Hearing, 24, 48–70.

                                                                              Bharadwaj & Graves: Efficacy of the DOVC Measure   635

                                        Downloaded from jslhr.asha.org on August 13, 2010
   Efficacy of the Discreteness of Voicing Category (DOVC) Measure for
Characterizing Voicing Errors in Children With Cochlear Implants: A Report

                  Sneha V. Bharadwaj, and Amanda G. Graves
                  J Speech Lang Hear Res 2008;51;629-635
                     DOI: 10.1044/1092-4388(2008/045)

              This information is current as of August 13, 2010

          This article, along with updated information and services, is
                       located on the World Wide Web at:

                  Downloaded from jslhr.asha.org on August 13, 2010

To top