Content uploaded by Raymond D Kent
Author content
All content in this area was uploaded by Raymond D Kent on Apr 17, 2021
Content may be subject to copyright.
clinical lingui stics & phonetics, 2001, vol. 15, no. 3,181± 193
The eŒects of ¯ attening fundamental
frequency contours on sentence intelligibility
in speakers with dysarthria
K AT E B U N T O N ² , R AY D . K E N T ² ,
J A N E F . K E N T² a n d JO S E P H R . DU F F Y ³
² Waisman Center, University of Wisconsin-Madison, USA
³ Mayo Clinic, Rochester, MN, USA
(Received 15 April 2000; accepted 7 August 2000 )
Abstract
This study explored the relationship between F0 variability and intelligibility in
persons with motor speech disorders. The literature suggests that monopitch may
have detrimental eŒects on the intelligibility of individuals with dysarthria (and
other speech disorders), but few studies have examined the independent eŒect of
a ¯ at F0 contour on intelligibility in speakers with articulatory di culties. An
LPC resynth esis technique was used to reduce the speakers’ sentence F0 range
by 25%, 50% and 100% (¯ attened F0) . Two dysarthric speaker groups were
evaluated, one with prosodic inadequacy and imprecise articulation (hypokinetic)
and one with imprecise articulation only ( U UMN ) . Speech intelligibility was
assessed using a word transcription task and an interval-scaling task. Results
argue strongly for the perceptu al importance of sentence-level F0 variations on
speech intelligibility even when the F0 range is severely restricted. Results also
suggest that the contribution of F0 to intelligibility may vary with type of
dysarthria.
Keywords: Dysarthria, prosody, intelligibility.
Introduction
The term `monopitch’ is frequently used to describe certain types of dysarthria
(Darley et al., 1969a, b). The lack of F0 variability across an utterance (the presumed
physical basis of this term) has been substantiated by several acoustic studies
(Yorkston and Buekelman, 1981; LeDorze et al., 1992; Bunton et al., 2000) .
However, relatively little attention has been focused on how much the perceptual
Address correspondence to: Ray Kent, University of Wisconsin-Madison, 435 Waisman
Center, 1500 Highland Avenue, Madison, WI 53705 , USA. e-mail: kent@waisman.wisc.edu
Clinical Linguistics & Phonetics
ISSN 0269-920 6 print/ISSN 1464-5076 online Ñ2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/02699200010003378
K. Bunton et al.182
system utilizes information supplied by prosodic patterns and its relationship with
speech intelligibility. Kent and Rosenbek (1982 ) reported that ¯ attened F0 contours
blur the contrast between adjacent units in the speech signal and have a detrimental
eŒect on intelligibility. However, they also noted that intelligibility and prosodic
disturbance are not necessarily equally impaired in all subjects.
Variation of F0 has been reported to contribute to intelligibility of neurologically
normal speech. Prosody has been shown to be an important perceptual cue for
vowel identity (Traunmuller, 1981 ) , stop consonant voicing (Haggard et al., 1970 ),
syllable stress ( Lehiste, 1970), and marking lexical boundaries (Liss et al., 1998 ).
Bradlow et al. ( 1996 ) studied both global and ® ne-grained diŒerences among talkers
as these diŒerences correlated with inter-speaker diŒerences in intelligibility. They
concluded that a highly intelligible speaker was one who produces sentences with a
relatively wide range of F0, a relatively expande d vowel space that includes a
substantial F1 variation, precise articulation of the point vowels and a high precision
of inter-segmenta l timing. Two studies that examined the independent eŒects of F0
contour on intelligibility in utterances produced by persons with no history of speech
disorders showed signi® cant diŒerences in intelligibility between the original utter-
ances and those with ¯ attened F0 contours (Laures and Weismer, 1999; Laures
et al., 1999 ). This decrease in intelligibility was found for subjects whose articulatory
precision was intact. For speakers with neurologic disease, who experience articulat-
ory di culties, there is little published informatio n about the independent relation-
ship between reduced F0 variation and speech intelligibility (for a review see Ramig,
1992 ). In developing an acoustic-phonetic model of intelligibility in dysarthria ( Kent
et al., 1989), it is of interest to know how much sentence-level prosody contributes
to speech intelligibility de® cits, in the presence of segmental-leve l articulatory de® cits.
The present study was designed to determine how a lack of sentence-level F0
¯ uctuation contribute s to speech intelligibility de® cits in two groups of speakers
with dysarthria. Speakers with hypokinetic dysarthria as the result of Parkinson ’ s
Disease typically exhibit decreased F0 ranges along with some segmental-level articu-
latory de® cits. Alternatively, speakers who experienced a cerebrovascular accident
resulting in unilateral upper motor neuron dysarthria typically have imprecise
consonants but do not exhibit di culties with prosody.
Methods
Speakers
Ten sentences randomly selected from the Fisher-Logemann Test of Articulation
Competence (1971), each 5± 7 words in length, collected from four speakers diagnosed
with Parkinson’s Disease (PD: 2M, 2F ), four with left cerebrovascula r accident
(LCVA: 2M, 2F ) , and two with right CVA (1M, 1F ) were analysed and used as
perceptual stimuli in the present study. The set of ten sentences was diŒerent for
each speaker, and was collected as part of a larger database on speech intelligibility.
Speaker characteristic s can be found in table 1. The range of intelligibility scores,
calculated from the Kent et al. ( 1989 ) intelligibility test, ranged from 85% to 96%.
This range of intelligibility was selected based on previous work demonstratin g that
the habitual amount of F0 variability decreases as overall intelligibility scores
decrease ( Bunton et al., 2000). The type of dysarthria was identi® ed by an experi-
enced speech pathologist (ibid.). These two types of dysarthria ( hypokinetic and
Flattening f undamental frequency contours 183
Table 1. Speaker characteristics. The type of dysarthria is based on judgement s of the Wnal
author who is a certiWed speech pathologist. The duration column includes time between
diagnosis and recording as well as length of time their speech has been aŒected if
diŒerent from medical diagnosis
Intelligibility
Subject Gender Age Diagnosis Duration Dysarthria mean (SD)
CS M 69 R CVA 27 days UUMN 86.7 (5.23)
TD F 23 R CVA 11 days UUMN 93.4 (2.7 )
FL F 78 L CVA 12 days UUMN 87.5 (1.6 )
BL M 71 L CVA 9 days UUMN 92.9
MB F 66 L CVA 18 days UUMN 85.46 (2.6 )
BT M 56 L CVA 8 days UUMN 86.52
EB F 75 P D 4 y speech/8 y Dx hypokin etic 92.52
LW F 61 P D 10 y speech/15 y Dx hypokinetic 95.25
LA M 70 PD 16 y hypokinetic 92.63
LB M 68 PD 1 y hypokinetic 88.95
unilateral upper motor neuron) were selected based on the perceptual prominence
of suprasegmental characteristics associate d with one but not the other, while both
dysarthrias are characterized by articulatory imprecision. Ten healthy, age-matched
speakers (® ve male, ® ve female) were included as controls.
Acoustic analysis
To quantify the habitual use of frequency variation for each speaker, the F0 minimum
and maximum were recorded for each utterance using Cspeech ( Milenkovic, 1997 ).
In addition, peak frequencies associated with each vocalic segment in the utterance
were measured, to ensure that the syllable structur e of the utterance was unchanged
during resynthesis of the utterances (see below).
Resynthesis
A linear predictive coding ( LPC) based algorithm that allowed for sentence-level
F0 modi® cations and did not aŒect the temporal or formant frequency characteristic s
of the utterances was used for resynthesis in the present study ( Milenkovic, 1999 ).
Each utterance was resynthesized prior to modifying the F0 contour. These unmodi-
® ed, resynthesized utterance s were used as control items in the listening experiments
to ensure that the technique itself did not have any eŒect on the intelligibility of the
sentences. The F0 contour of the utterance was then modi® ed in three ways: (1 ) each
voiced segment was ¯ attened by setting all pitch periods equal to the mean value
calculated over the entire utterance (hereafter referred to as ¯ at F0); (2) the range
of F0 variability was reduced by 25% while maintaining the syllable structur e of the
utterance; and (3) the range was reduced by 50%. For conditions 2 and 3, peak F0
for each vocalic segment was subtracted from the mean. This value was multiplied
by the percentage decrease (25% or 50%) and the resultant value was added to
original peak value, thus bringing the peak 25% (or 50%) closer to the mean value.
For example, if the F0 mean for an utterance was 150 Hz and the individual peak
was 200 Hz and a reduction of the F0 range by 25% was desired, the following
equation applies [( 150Õ200)Ö0.251200 5187.5 Hz] . This procedure was used for
K. Bunton et al.184
each individual peak within the utterance thereby preserving the F0 contour of the
speaker’ s original production. An example of a spectrogram and F0 contour for an
original utterance (panel a) an unmodi® ed, resynthesized utteranc e (panel b ), one
with an F0 range reduced by 25% ( panel c), and one with a ¯ at F0 ( panel d ) are
shown in ® gure 1 to illustrate that the resynthesis technique did not modify the
spectral and tempora l properties of the utterance. The utterance shown is `They all
know what I said’ produced by a NG female speaker. The 800 utterances ( ten
utterances per speakerÖfour resynthesis conditionsÖ20 speakers) were placed into
a single list to be used during the perceptual portion of the experiment.
Listeners
Ten graduate students ( 5M, 5F ) served as listeners. All students had taken a graduate-
level dysarthria course and had no self-reported hearing loss. Listeners with limited
exposure to dysarthric speech were selected to avoid ceiling eŒects related to either
the novelty of the speech or highly tuned listening skills. Listeners were seated in a
sound booth and stimuli were presented over a loudspeaker at a comfortable listening
level. The order of the utterances from the 800-utterance pool was generated randomly
by the computer for each listener to control for order eŒects. Speakers heard all 800
utterances for each of the listening tasks as well as 40 repeat utterances ( ten from
each resynthesis condition) to obtain a measure of intra-judge reliability. The listening
task was divided into four sections of 420 utterances each and listeners were given a
20-minute break between sections. The ® rst time the listeners heard the utterances
(sets 1 and 2), they were asked to write down what they heard as accurately as
possible. The second time the listeners heard the utterances (sets 3 and 4) they were
asked to assign a scale value of intelligibility using a 7-point equal-interval scale. The
scale was de® ned for listeners as `A scaling of 1 is equal to 0% intelligible; a scaling
of 7 is equal to 100% intelligible. For this study, intelligibility is de® ned as the ease
with which you understand the words spoken’.
Intelligibility measures
The purpose in intelligibility testing in the present study was twofold. The ® rst
measure of intelligibility, transcription, was selected to get a measure of the degree
to which the listener recovers the discriminations intended by the speaker, that is,
to determine the accuracy with which the speaker’s intended message was recovered
by the listener. The second measure, scaling, provided a quick, overall index of a
speaker’ s intelligibility based on the listeners’ perceived eŒort in recovering the
message. Segmental level errors produced by the speakers were not modi® ed during
resynthesis of the F0 contours , therefore, an explanatory approach to intelligibility
testing designed to identify the phonetic loci of the intelligibility de® cit such as
presented by Kent et al. ( 1989, 1990, 1992) was not appropriate for the present study.
Data analysis
Data analysis for sets 1 and 2 were achieved by calculating the number of words
correctly transcribed for each utterance and speaker, and averaging across listeners.
Scaled values (sets 3 and 4) were also averaged across listeners. Non-parametric
statistics were used to separatel y analyse the results of the transcriptio n and scal-
ing tasks.
Flattening f undamental frequency contours 185
Figure 1. Examples of spectrograms and F0 contours the token `They all know what I said’.
Panel a is the original utt erance, followed by a resynthesized, unmodiWed condition
(panel b), 25% reduced F0 range (c), and Xat F0 (d) produced by a NG female.
Reliability
Forty utterances were repeated from the set of 800 original tokens for each listener
to obtain measures of intra-judge reliability. The overall correlation coe cients
K. Bunton et al.186
between the ® rst and second presentation of each utteranc e ranged between 0.83
and 0.97 for the ten listeners on the transcriptio n task. For measures of inter-judge
reliability, the correlation coe cients were between 0.77 and 0.98 for each speaker
across listeners. For the scaling task, reliability was considerably lower for both
inter- and intra-judge measures. The inter-judge coe cients ranged from 0.62 to
0.88 and the intra-judge values ranged from 0.54 to 0.78. No systematic shifts ( i.e.,
improvements) in performance were noted for individual listeners across tokens.
Results
Acoustic characteristics of speakers
Table 2 shows the means and standard deviations for the mean F0, mean high and
low F0 and the range of F0 variability averaged across the ten utterance s for each
speaker. To calculate the range of F0 variation, the high and low frequency peaks
for each utterance were measured and values were then averaged across the utter-
ances. This method was used rather than subtracting the mean high and low F0
values to account for the variability in performance for individual speakers. The
range of F0 across utterances for all disordered speakers was smaller than for the
control speakers (see also Bunton et al., 2000 ) . The reduction in range varied from
22% to 68%. A compariso n of range of F0 variability with intelligibility scores failed
to show a relationship between F0 variability within an utterance and the overall
intelligibility score. In fact, the two PD speakers with the highest intelligibility scores
have the most restricted frequency ranges. This data can be seen in ® gure 2. In this
® gure the NG speakers are represented by diamonds, the PD speaker s by squares,
R CVA by triangles and L CVA by circles.
Table 2. Mean and standard deviations shown in parentheses for the mean, high, low, the range
of F0 values for each speaker
Speaker Gender Dx Mean F0 High F0 Low F0 F0 range
CS M R CVA 127.6 ( 4.3 ) 181.4 (14.8) 101.5 ( 5.9 ) 72.3 (13.3)
TD F R CVA 179.2 ( 22.3 ) 213.5 (15.4 ) 142.8 ( 8.9 ) 79.0 (16.4 )
FL F L CVA 186.5 ( 12.4 ) 242.8 ( 6.5 ) 164.7 (9.4 ) 70.5 ( 18.4 )
BL M L CVA 123.1 (16.6) 156.3 ( 8.4 ) 114.5 (5.4 ) 36.0 ( 17.2 )
MB F L CVA 158.8 (9.7) 186.9 ( 4.7 ) 135.8 (7.4 ) 55.3 ( 18.4 )
BT M L CVA 118.5 (9.8) 164.2 (13.5 ) 110.0 ( 10.6 ) 58.6 (13.2 )
EB F PD 215.4 ( 16.8 ) 240.5 ( 10.2 ) 192.5 ( 6.8 ) 57.2 (17.1)
LW F PD 199.7 ( 18.4 ) 215.5 (12.5 ) 181.4 ( 8.1 ) 45.0 (6.9 )
LA M PD 117.5 (13.1) 146.8 ( 11.4 ) 109.8 ( 10.0 ) 26.7 ( 16.1 )
LB M PD 146.5 ( 9.8) 168.2 (9.7) 135.3 (4.1) 59.2 (14.6 )
NG1 F 231.5 ( 16.1 ) 261.5 (13.2 ) 166.2 ( 16.8 ) 110.5 ( 16.4 )
NG2 F 202.8 ( 11.4 ) 260.8 (16.7 ) 154.8 ( 16.7 ) 125.7 ( 17.2 )
NG3 F 192.8 ( 21.5 ) 245.1 (12.5 ) 164.2 ( 21.7 ) 116.7 ( 13.7 )
NG4 F 215.7 ( 5.1 ) 269.4 (5.2 ) 141.1 (18.4 ) 132.5 (11.8)
NG5 F 216.7 ( 7.8 ) 277.6 (6.5 ) 175.2 (26.4 ) 109.5 (16.2)
NG6 M 126.8 (15.2) 188.2 ( 6.4 ) 90.5 (16.4 ) 105.2 (17.4)
NG7 M 158.7 (16.2) 190.9 ( 13.7 ) 110.0 (14.3) 89.4 (13.9 )
NG8 M 148.9 (9.8) 196.2 (13.4 ) 92.5 (12.6 ) 110.5 (17.2 )
NG9 M 151.0 (13.4) 199.8 ( 9.5 ) 99.9 (13.7 ) 93.8 (11.7 )
NG10 M 133.9 (21.1) 179.5 ( 13.6 ) 85.7 (6.9 ) 98.7 ( 13.4 )
Flattening f undamental frequency contours 187
Figure 2. Speech intelligibility scores plotted against natural F0 range for the three disorder
groups and mean values for the NG male and female speakers (diamonds5NG,
squares 5PD, triangles 5RCVA, circles 5LCVA).
Transcription data
Transcription scores were calculated by counting the number of words correctly
recorded out of a total of 70. Results of the perceptual transcription task showed
an expected decline in transcriptio n score corresponding to the systematic decrease
in the F0 range for the NG and CVA speakers (® gure 3), although for the NG-M
group the reduction in intelligibility was negligible. The PD speakers, on the other
hand, showed a decrease in accuracy for the habitual utterances and 25% and 50%
reduced F0 range utterances. The utterances with a ¯ attened F0 contour, however,
had higher transcription scores than those utterances resynthesized with reduced
ranges. The ¯ at F0 transcription scores were comparable to the speakers’ habitual
utterances. Plots of the frequency range and transcription scores are shown in
® gure 4 for individual speakers: the three panels of the ® gure each represent speakers
with a diŒerent neurologica l diagnosis. In these panels the female speakers are shown
as triangles and the male speakers as circles. The PD speakers (panel a) all showed
an increase in transcriptio n scores for ¯ at F0 utterances compared to the tokens
Figure 3. Means and standard deviations on the transcription task for each speaker group
across listeners. The total number of words possible was 70.
K. Bunton et al.188
Figure 4. Transcription scores plotted against F0 variation for individual speakers in the three
disorder groups.
with reduced F0 ranges. Results of non-parametri c statistica l analysis show signi® c-
ant diŒerences between habitual utterances and range-reduced utterances for the PD
speakers and between the habitual utterances and only ¯ at F0 utterances for the
other three speaker groups (table 3).
Scaling data
Results of the scaling task were similar to those found for the transcription task
(® gure 5) , the listeners rated utterance s with decreased or ¯ at F0 contours as less
intelligible or more di cult with scaling scores closer to 1. This is consistent with
the decline in transcriptio n scores, with the exception of the PD speakers. For the
two female PD speakers there was little change in their scaling scores across the
four resynthesis conditions, whereas for the male PD speakers there was a drastic
fall in the scaling scores when F0 contours were reduced. This occurred despite
Flattening f undamental frequency contours 189
Table 3. Results of non-parametric statistical analysis comparing the habitual utterances with
the modiWed ones. A letter indicates a signiWcant Wnding (T 5transcription, S 5scaling)
Speaker Group 25% reduction 50% r eduction Flat F0
NG M T
F T
PD M T S T
F T T
RCVA M T
F T
LCVA M T
F T
Figure 5. Mean scaling scores for each speaker group across listeners. A 7.0 Lickert scale was
used for this task.
improvement in transcriptio n scores when the F0 contour was ¯ attened (® gure 4).
Plots of the F0 range versus scaling scores for individual speakers are shown
in ® gure 6. Similar to the transcriptio n plots, female speakers are represented as
triangles and male speakers as circles.
Discussion
This study was designed to explore the relationship between F0 variability and
intelligibility in persons with hypokineti c and UUMN dysarthria associated with
PD and CVA, respectively. Results argue strongly for the perceptual importance of
sentence-level F0 variations on speech intelligibility in the presence of segmental-
level errors for both speaker groups. It appears that even though reduced F0 ranges
were characteristic of the habitual utterances produced by the dysarthric speakers
in the present study (table 2), the frequency information was an important perceptual
component of the signal. Further ¯ attening the F0 range using a resynthesis technique
resulted in signi® cant decreases in intelligibility for both groups as well as in the
control speakers. In the control group decreased intelligibility scores were also seen
as the F0 was compressed, however, the eŒect on overall sentence intelligibility was
roughly half that of the disorder groups (® gure 3). The mean decrease in transcriptio n
K. Bunton et al.190
Figure 6. Scaling scores plotted against F0 variation for the three disorder groups.
score for the NG-M and NG-F groups were 8.6 and 16.4 words, respectively,
whereas the decrease for the CVA groups were between 21.7 and 27.1 words.
Comparing the right and left CVA groups, lower transcription scores were found
for left versus right although the diŒerence was not statistically signi® cant.
One question guiding this experiment was whether or not there was a critical
amount of F0 variability below which speech intelligibility would decrease precipit-
ously. Looking at the ® ndings for the NG speakers in the present study, it does not
appear that there is such a value. A continuous and gradual decrease in transcription
scores was found as the frequency range was reduced. The same conclusion can be
drawn for the CVA speaker groups; a systematic decline in transcription scores,
which corresponded to the decreasing F0 range, was found even when the speakers
exhibited segmental-level compromise that may have increased the di culty of the
task for the listener. Results for the PD speakers, on the other hand, were peculiar.
These speakers showed a similar decrease in transcription scores compared to normal
and CVA speakers, but when F0 informatio n was removed entirely from the signal
Flattening f undamental frequency contours 191
their transcription scores improved to nearly what they were for their habitual
productions (® gure 4) . One possible explanation for this ® nding could be that the
arti® cially restricted frequency variability (seen in the 25% and 50% range reduction
resynthesis) provided con¯ icting or ambiguous information within the signal increas-
ing the listeners di culty in understanding the sentence. This seems consistent with
the lower scaling scores found during the second part of the experiment, especially
for the two male PD speakers (® gure 5). Another possible explanation is that because
PD is a degenerative disease these speakers have adopted alternative or compensatory
strategies to preserve speech intelligibility in the presence of other system limitations
(e.g., ability to modulate F0). In the present case, temporal cues seem to be the
most logical candidate and further investigation is warranted. In the utterances
which did not contain any frequency information, perhaps listeners were able to
successfully alter their perceptual strategies and use these `exaggerated’ duration
cues to aid in their understandin g of the utterances. For the utterances with arti® cially
limited F0 variability (resynthesized with 25% or 50% range reduction) there may
have been a mismatch between F0 and temporal information which resulted in lower
transcription scores. Further analysis of a second dysarthria type, also the result of
a progressive neurologic disease (e.g. ALS ), may shed light on whether and how
compensatory strategies are being used by these speakers to maintain speech
intelligibility.
Results of the scaling task were consistent with the transcription task: reduction
in the range of F0 variability corresponde d to lower scaling scores re¯ ective of
increased di culty or eŒort reported by the listeners. A regression analysis between
the transcription scores and the scaling scores, however, showed a fairly strong
relationship between the two tasks (® gure 7). However, low values of inter- and
intra-judge reliability raises questions about the ability of the listeners to perform
this task and thus limit conclusions that can be drawn from these results.
It was hypothesized in the present study that in dysarthric speech, where acoustic
representations of supraglottic articulation are likely to be de® cient, the eŒect of
¯ attening an F0 contour on speech intelligibility would be greater than for control
Figure 7. Regression analysis for the transcription score versus scaling score across speakers.
K. Bunton et al.192
speakers who show no articulatory di culties. Both the neurologically normal
speakers and those with neurologic disease showed a decrease in transcriptio n and
scaling scores as the F0 range was reduced, however, the eŒect was larger for the
speakers whose articulatory precision was compromised. Further, no critical amount
of F0 variability was found, and it appears that syllable contrast resulting from even
minor F0 variations is an important perceptual cue, thus contributing greatly to
speech intelligibility.
Acknowledgements
This work has been funded by NIH R01 DC00319 and T32 DC00042. Parts of this
work were included in a presentatio n made at the 10th Biennial Conference on
Motor Speech, 2000.
References
Bradlow, A., Torretta, G. and Piso ni, D., 1996, Intelligibility of normal speech I. Global
and ® ne grained acoustic-phone tic talker characteristics. Speech Communication,20,
255± 272.
Bunton, K., Ken t, R., Kent, J. and Ro senbek, J., 2000, Perceptuo-acoustic ass essment of
prosodic impairment in dysarthria. Clinical Linguistics and Phonetics,14, 13± 24.
Darley, F., Aronson, A. and Brown, J., 1969a, DiŒere ntial diagnostic patterns of dysarthria.
Journal of Speech and Hearing Research,12, 246± 269.
Darley, F., Aronson, A. and Brown, J., 1969b, Clusters of deviant speech dimensions in
the dysarthrias. Journal of Speech and Hearing Research,12, 462± 496.
Fisher, H. B. and Logemann, J., 1971, Fisher-Logemann Test of Articulation Competence
(Boston: Houghton Mi‚ in).
Haggard , M., Ambler, S. and Callow, M., 1970, Pitch as a voicing cue. Journal of the
Acoustical Society of America,47, 613± 617.
Kent, J.F., Kent, R.D., Rosenbek, J., Wei smer, G., Martin, R., Sufit, R.and
Brooks, B. R., 1992, Quantitative description of the dysarthria in women with
amyotrophic lateral sclerosis. Journal of Speech and Hearing Research,35, 723± 733.
Kent, R.D., Kent, J.F., Weismer, G., Suf it, R., Rosenbek, J.C., Martin, R. E. and
Brooks, B. R., 1990, Impairment of speech intelligibility in men with amyotrophic
lateral sclerosis. Journal of Speech and Hearing Disorder s,55, 721± 728.
Kent, R., Weismer, G., Ken t, J. and Ro senbek, J., 1989, Toward phonetic intelligibility
testing in dysarthria. Journal of Speech and Hearing Disorders,54, 482± 489.
Kent, R. and Rosen bek, J., 1982, Prosodic disturbance and neurologic lesion. Brain and
Language,15, 259± 291.
Laures, J. and Wei smer, G., 1999, The eŒect of a ¯ atte ned F0 on intelligibility at the sentence-
level. Journal of Speech, Language, and Hearing Research,42, 1148± 1156.
Laures, J., Bu nton, K. and Weismer, G., 1999, The eŒec t of Xattened F0 contours on sentence
intelligibility. Paper presented at the 1999 meeting of the American Speech, Language
and Hearing Association, San Francisco, CA.
LeDorze, G., Dio nne, L., Ryalls, J., Julien, M. and Ouellett, L., 1992, The eŒect of
speech and language therapy for a case of dysarthria associated with Parkinso n’s
disease. European Journal of Disorders of Communication,27, 313± 324.
Lehiste, I., 1970, Suprasegmentals (Cambridge, MA: MIT Press).
Liss, J., Spitzer, S., Caviness, J., Adler, C. and Edwards, B., 1998, Syllabic strength and
lexical boundary decisions in the perception of hypokinetic dysarthric speech. Journal
of the Acoustical Society of America,104, 2457± 2466.
Milenkovic, P., 1997, Cspeech Version 4.0 (computer program: University of Wisconsin-
Madison).
Flattening f undamental frequency contours 193
Ramig, L., 1992, The role of phonation in speech intelligibility: a review and preliminary
data from patients with Parkinson’s Disease. In R. D. Kent (Ed.), Intelligibility in
speech disorders ( Amsterdam: John Benhamins), pp. 119± 155.
Traunmuller, H., 1981, Perceptual dimensions of openn ess in vowels. Journal of the
Acoustical Society of America,69, 1465± 1475.
Yorkston, K. and Bu ekelman, D., 1981, Ataxic Dysarthria: treatment sequences based on
intelligibility and prosodic considerations. Journal of Speech and Hearing Disorders,
46, 398± 404.
