Structured Variation across Sound Contrasts, Talkers, and...
Transcript of Structured Variation across Sound Contrasts, Talkers, and...
§ Speech is variable due to talkers, contexts, speaking styles, etc.§ Individual talker variability observed for;
· Vowels (e.g. Peterson & Barney 1952, Johnson et al. 1993) · VOT (e.g. Allen et al., 2003; Scobbie, 2006, Chodoroff et al. 2015)· CoG (e.g. Newman et al. 2001)
BACKGROUND
§ Confederate-led scripted dialogues§ 32 participants (80 dialogues per participant) (Ohala, 1994; Lindbolm et al. 2007; Maniwa, Jongman & Wade, 2009)§ Measurements (n=2514)
/ p, t / / s, ʃ /
VOT CoGduration
Locus equation (LE) slope (co-articulation)Vowel dispersion V = /i/, /æ/, /ɑ/, /oʊ/, or /u/
(Euclidean distance Neary normlalized)
Speaking rate (num. of syllables per sec.)
§ Prominence-induced “Clear speech” by confederate “mishearing” portion of dialogue · Control condition: TARGET heard correctly
· Prominent condition: TARGE misheard for another C (p,t,s or ʃ)
§ How systematic are talker differences? · Are they stable across consonants? § How does this relate to speech style? · Do talkers differ in how clear they are?
QUESTIONS
METHODS SAMPLE DIALOGUES
RESULTS
This research was supported by SSHRC grant # 435-2014-1504 to Meghan Clayards and Michael Wagner. The authors would like to thank Alissa Azzimmaturo for help with data collection and Charlene Alcena and Sarah Perillo for help with annotation.
References: Allen, Miller, & DeSteno (2003). Individual talker differences in voice-onset-time. JASA, 113(1), 544-522./ Chodroff, Godfrey, Khudanpur, & Wilson (2015). Structured variability in acoustic realization: A corpus study of voice onset time in American English stops. Proceedings of ICPhS Glasgow, UK. / Johnson, Ladefoged, & Lindau (1993). Individual differences in vowel production. JASA 94(2), 701-714. / Krull (1988). Acoustic properties as predictors of perceptual responses: A study of Swedish voiced stops. Phonetic Experimental Research, Institute of Linguistics, University of Stockholm (PERILUS) 7. / Lindblom (1990). Explaining phonetic variation: A sketch of the H&H theory. In W. J. Hardcastle & A. Marchal (Eds.), Speech Production and Speech Modeling (pp. 403-439), Dordrecht:Kluwer. / Lindblom, et al. (2007). The effect of emphatic stress on consonant vowel coarticulation. JASA, 121(6), 3802- 3813. / Newman, Clouse, & Burnham (2001). The perceptual consequences of within-talker variability in fricative production. JASA, 109(3), 1181-1196. / Recasens, Pallarès, & Fontdevila, (1997). A model of lingual coarticulation based on articulatory constraints. JASA, 102(1), 544–561. / Theodore, R. M., Miller, J. L., & DeSteno, D. (2009). Individual talker differences in voice-onset-time: Contextual influences. JASA, 125(6), 3974-3982 / Scobbie, J. M. (2006). Flexibility in the face of incompatible English VOT systems. In: Laboratory Phonology 8 Varieties of Phonological Competence. Phonology and Phonetics 4-2 . Mouton de Gruyter, Berlin, pp. 367-392.
Structured Variation across Sound Contrasts, Talkers, and Speech Styles Hye-Young Bang1 & Meghan Clayards12
[email protected] , [email protected] of Linguistics, McGill University,
2School of Communication Sciences and Disorders, McGill University
§ Talkers differ systematically across sound contrasts within a speaking style§ Talkers differ in degree of hyper-articulation§ Correlations between cues across speakers tend to be strengthened in clear speech (under prominence)§ Differences between talkers (at least partly) reflect global properties of talkers§ The systematic differences in talkers may help listeners quickly adapt to talker variability in speech perception
CONCLUSION
LabPhon 15Ithaca, NY
July 13-16, 2016
CO
NTR
OL
PRO
MIN
ENT
Have you heard of “grey peep?”
Green peep?No, GREY peep.
What? Great peep?No! GREY peep.
CONFEDERATE PARTICIPANT
Have you heard of “grey peep?”
Grey what?Grey PEEP.
What? Grey teep?No! Grey PEEP.
0.4
0.5
0.6
0.7
100 150
VOT(/p t/)
/s/_
vowe
l disp
ersio
n
prom. no yes
(β = 0.32, p = 0.027)0.4
0.5
0.6
0.7
150 200 250 300
/s/_duration
stop
_vow
el d
isper
sion
prom. no yes
(β = 0.42, p = 0.012)
6000
8000
10000
100 200 300 400
/s/_duration (mean)
/s/_C
oG
prom. no yes
(β = 0.34, p = 0.058)
100
200
100 200 300 400
/s/_duration (mean)
VOT(
/p t/)
prom. no yes
(β = 0.76, p <.001)
0.3
0.4
0.5
0.6
0.7
150 200 250 300
/s/_duration
/s/_
LE S
lope
prom. no yes
(β = -0.37, p = 0.048)
(r= 0.76, p <.001)
Differences in body size or gender?
Differences in speaking style?
0.4
0.5
0.6
0.7
100 150
VOT(/p t/)
stop
_vow
el d
ispe
rsio
n
prom. no yes
(β = 0.37, p = 0.021)
4000
6000
8000
10000
12000
50 100 150 200 250
VOT(/p t/)
/s/_
CoG
prom. no yes
(β = 0.42, p = 0.013)
§ Stats are from ”prominent condition” with speaking rate as a control factor in regression
§ VOT correlated with /s/ CoG§ /s/ duration correlated with /s/ CoG § VOT correlated with /s/ duration§ VOT and /s/ duration correlated with vowel
dispersion across consonantslonger /s/ = smaller slope =less co-articulation
4500
5000
5500
6000
6500
7000
100 125 150 175 200
Mean F0
Mea
n Co
G
60
80
100
120
155cm158cm
160cm163cm
165cm168cm
170cm173cm
175cm185cm
188cm
height(cm)
Mea
n VO
T
0
50100150200
no yes
prominence
VOT
100
200
300
no yes
prominence
/s/_duration
4000
6000
8000
10000
no yes
prominence
CoG
1500
2000
2500
1000150
0200
0250
0300
0
F2_mid
F2_onset
m f 4500
5000
5500
6000
6500
7000
155cm158cm
160cm163cm
165cm168cm
170cm173cm
175cm185cm
188cm
height(cm)
Mea
n C
oG
60
80
100
120
100 125 150 175 200
Mean f0
Mea
n VO
T
same pattern as between talkers
§ Clear speech · VOT↑ · /s/ duration↑ · CoG↑ · LE slope↓ · vowel dispersion↑