The relationship between sonority and glottal vibration - …linguistics.ucla.edu/people/keating/Risdal-etal_CUNY2016_final.pdfThe relationship between sonority and glottal ... Introduction

The relationship between sonority and glottalvibration

Megan L. Risdal, Ann Aly, Adam J. Chong,Pat Keating, Jesse Zymet

CUNY Phonology Forum — 2016 Sonority

Introduction & Background Our Study Results Summary Acknowledgments References

Introduction

I Sonority and the sonority hierarchy have been an importantbasis underlying explanations for various phonologicalphenomena (e.g., assimilation, phonotactics of syllablestructure).

Sonority & Glottal Vibration — CUNY 2016 January 19, 2016— Slide 2/32


Phonetics of Sonority

Searching for a Physical Basis of Sonority

I How it manifests phonetically and how to quantify it iscurrently unresolved (Eduard, 1881; Heffner, 1950; Fletcher,1972; Parker, 2002).

I Some have suggested that sonority may be related to:I degree of constriction (e.g., Hankamer and Aissen, 1974)

I While others have related sonority to:I voicing (e.g., Nathan, 1989)




Two Scales of Sonority

I Miller (2012) uses both of these parameters:I Source scale: No source < turbulence only < breathy voice

(periodic source + glottal source) < modal voice (periodicsource)

I Aperture scale: explosive stops < voiced implosives <fricatives < nasals < liquids and glides < vowels

I The two parameters are considered to be completely separate.

I The two scales are combined additively to determine asegment’s sonority.




Phonetic Correlates of Sonority

I Parker (2002) also took into account both voicing and degreeof constriction in several of the phonetic correlates hemeasured (e.g. intensity, peak intraoral air pressure and peaktotal air flow) which depend simultaneously on both sourceand filter.

I Other measures, F1 and duration, depend only on the filter.

I He reports that intensity exhibits the strongest (positive)correlation with the sonority hierarchy and duration exhibitsthe least (see also Gordon, this conference).



Glottal Source-Vocal Tract Interactions


I Voicing requires airflow through the glottis.

I With no oral constriction→ air flows freely and voicing is easy.

I With a complete oral constriction → glottal airflow stops andvoicing stops (Ohala, 1983; Westbury and Keating, 1986).

I In between these two extremes, voicing should vary dep. onthe degree of the oral constriction.

I It has also been suggested that the way the vocal folds vibrateprobably varies, too.




Previous evidence: effects of oral constriction on voicing

I Voicing amplitude varies (Sole, 2015).

I Glottal excitation (rate of glottal closing) varies (Fant, 1997;Stevens, 2000; Bickley and Stevens, 1987).

I Degree of glottal constriction (amount of glottal flow) varies(Fant, 1997; Stevens, 2000; Bickley and Stevens, 1987).

I Suggests that tighter oral constriction → breathier voicing.



Mittal et al.

Mittal et al. (2014)

I Six geminate voiced consonants: [r, z, G, l, n, and N].

I Measured strength of glottal excitation ∆ψ:An acoustic measure of the relative amplitude of impulse-likeexcitation derived from the instant of significant excitation ofthe vocal-tract system during production of speech (Murtyand Yegnanarayana, 2008).

I Result: [z] < [r] < [n,N] < [G] < [l].


Our Study


Our Study

Our Study: Questions

I Replicate and extend Mittal et al. (2014) with moreconsonants and adding vowels, focusing more explicitly on:

1. Does voicing vary across segment types, and if so;

2. Do these distinctions depend on degree and type ofconstriction?

3. Do these differences among voiced segment types correlatewith standard notions of sonority?



The Study Design

Segments (21 total)

I Including 5 of 6 segments from (Mittal et al., 2014). Weeliminated [N] because no difference between it and [n] wasreported.

Consonants

I Approximants: [j, w, l, ô]

I Trill & Tap: [r, R]

I Nasal: [n]

I Fricatives: [D, z, G, K]

I Affricates & Stop: [Ã, gG, d]

Vowels

I Front unround: [i, e, a]

I Front round: [y, ø]

I Back round: [o, u]



The Study Design

Our Dependent Measures

I Acoustic signal: Strength of glottal Excitation (SoE), thesame dependent measure, ∆ψ, reported in Mittal et al.(2014).

I Electroglottography (EGG) signal: the Contact Quotient(CQH) as a measure associated with amount of vocal foldcontact during phonation.



The Study Design

Electroglottography (EGG) & Contact Quotient

I Contact Quotient (CQH): the proportion of a completevibratory cycle from an inferred point of closure to an inferredpoint of opening.



The Study Design

Participants & Procedure

I 13 participants (6 males and 7 females) contributed data forthe present study.

I All speakers were linguists trained in phonetics and fluentspeakers of English (8 native speakers of North AmericanEnglish).

I Participants produced consonants (3 reps [aCa]) and vowels (3reps [wV]) for a total possible 63 tokens per speaker or 819overall.



The Study Design

Segmentation & Measurement

I Tokens were segmented using Praat TextGrids from onsetthrough voiced duration.

I Only fully-voiced portions of segments were analyzed.I The first voiced closure of the trill with at least 3 pulses was

segmented.I Affricates were treated as a voiced closure (stop) and a voiced

release (fricative) separately.I Stop releases were not included.

I Using VoiceSauce for SoE (Shue et al., 2011) and EggWorksfor CQH (Tehrani, 2012) software.

I Missing values and outliers were excluded leaving 744 tokensfor analysis. We report on speaker normalized values of CQH

and SoE.


Results & Analysis


Descriptives & Analysis Method

Missing Tokens: Evidence for Voicing Difficulty

Figure: Segments eliminated due to lack of voicing. L −→ R =decreasing constriction degree.



Descriptives & Analysis Method

Analyses

I First we examine overall trends between CQH & SoE andvoiced segments and discuss our findings as they relate tosonority.

I Following this, we take a closer look at several within-categorydistinctions.

I Null hypothesis: no difference in CQH or SoE across voicedsegments, with the exception of voiced fricatives.


CQH : Consonants

Figure: CQH across all consonants, collapsing vowels into a singlecategory. L −→ R = decreasing constriction degree.

Strength of Excitation: Consonants

Figure: SoE across all consonants, collapsing vowels into a singlecategory. L −→ R = decreasing constriction degree.

The relationship between CQH and SoEI CQH makes distinctions among obstruents and SoE among

sonorants.

ɹ

ɣʁ ɾ

a

ð

d_cl

e

gɣ_cl

gɣ_rel

i

j

ln

o

ø

r

ʤ_rel

u

w

y

z-1.0

-0.5

0.0

0.5

1.0

1.5

-1.0 -0.5 0.0 0.5Mean Scaled Contact Quotient

Mea

n S

cale

d S

oEmannera

a

a

a

a

a

a

a

stopfricativetrilltapnasalliquidglidevocalic

Figure: Scaled CQH by Scaled SoE by segment. Size of symbol indicatesstandard deviation.


Results: CQH and SoE

Conditional Inference Tree

I Do CQH values discriminate among voiced sounds along thesonority hierarchy?

I Conditional inference trees work by recursively partitioningobservations into subsets of the data which are significantlydifferent with respect to an outcome variable (Hothorn et al.,2006).


CQH : All SegmentsI V > liquids/glides + [gG] (cl) > non-dorsal fricatives/tap/trill

+ [d] (cl) > dorsal fricatives

segmentsp < 0.001

gɣ_cl, n, l, ɹ, j, w, V d_cl, gɣ_rel, ʁ, ɣ, ʤ_rel, ð, z, r, ɾ

segmentsp < 0.001

gɣ_cl, n, l, ɹ, j, wV

Node #2 (n = 201)

-3

-2

-1

0

1

2

Node #1 (n = 266)

-3

-2

-1

0

1

2

segmentsp < 0.001

d_cl, gɣ_rel, ʤ_rel, ð, z, r, ɾ , ʁ ɣ

Node #3 (n = 213)

-3

-2

-1

0

1

2

Node #4 (n = 50)

-3

-2

-1

0

1

2

Figure: CQH across all segments.


Results: CQH and SoE

Interim Summary: The relationship between CQH and SoE

I A range of values of both measures is observed acrosssegments of different constriction degrees.

I In broad terms, the trends reflect differences along thesonority hierarchy.

I Vowels, on the whole, have the highest CQH and SoE values,indicating strongest voicing energy and least breathy voicing.

I Compared to nasals, liquids and glides, voiced obstruents alsohave lower CQH and SoE values - less energy in voicing andbreathier voicing.



Results: Within Segment Types

Analyses: Within Segment Types

I Cross-linguistically, the sonority hierarchy is stable at eachlevel; however, there is some variation within levels (Janyet al., 2007).

I In particular, liquids and rhotics differed across languages (e.g.,Egyptian Arabic, Hindi, Mongolian, Malayalam).

I Other contrasts may pattern differently cross-linguistically, forexample sibilants versus other fricatives.

I This motivates us to examine some within-categorydifferences.

I Linear mixed effects models1 in R predicting CQH/SoE values.

1Speaker as random intercept. P-values obtained using likelihood ratio tests.Sonority & Glottal Vibration — CUNY 2016 January 19, 2016— Slide 25/32



Trill, tap, and [d]

I We were interested in examining segments that involved fullclosures, but are nonetheless in different parts of the sonorityhierarchy.

I The trill has a significantly lower CQH value (breathier)compared to [d] and [R] which aren’t significantly different.

I That is: [r] < [d] = [R] on CQH values.

I Trills pattern like voiced fricatives which is not surprising astheir contact intervals are often more like voiced fricativesthan like stops.




Place (dorsal/coronal) versus Sibilance

I Mittal et al. (2014) attributed a difference between dorsal,non-sibilant [G] and coronal, sibilant [z] in SoE to a place ofarticulation effect rather than an effect of sibilance.

I We compared the dorsal, non-sibilant fricatives [K] and [G](n = 59) to the coronal, non-sibilant [D] (n = 34) and thecoronal, sibilant [z] (n = 38) to tease this apart.

I Place: Dorsal fricatives (non-sibilants) have lower CQH

values than coronal [D] (non-sibilant) and [z] (sibilant) →breathier voicing.

I Sibilance: no significant effect on model fit.


Strength of Excitation: VowelsI We tested the relationship between voicing across vowels and

intrinsic aperture using scaled SoE and scaled F1 as anindicator of vowel height/aperture.

a

aa uu

uy

yy

øø

øo

o

o

i

iie

eeø

ø

ø

e

ee

i

ii

y

y

yu

uu

aaa

ooo y

yy

oooa

a

a

i

ii

uuu

øøø

e

e

e

aa

eee

iii

y

y

oo

o

u

uu

øøø

uu

uyyyø

ø

øee

e

aaa

ii

i

oo

o

i

e

a

øy

ou i

i

e

eaa

øø

yy

oo

u

u

ooo iiiaa a

ø

ø

ø

eee

uu

uy

yy

uuu

a aa

e

e

e

ii

i

ø ø

ø

o oo

øøø

aa

a

i

ii

ee

e

u

u

uo

oo

yy

a

uuuo

o

o i

ii

eee

y

yy

øøø

i

i

i

yy

yu

u

u

aaa

ee

e

ooo

aaa

yyy

o

oo ø

ø ø

uuu

ii

i

eee

eee

uuu

øøø

ii

i

yy

yo

oo

aaa

-1

0

1

2

3

-10123Scaled F1 (reversed)

Sca

led

Stre

ngth

of E

xcita

tion

vowelsaaaaaaa

yuiøoea

Figure: Strength of Excitation versus scaled F1 (reversed).



Vowels: [i, e, a, y, ø, o, u]

What were the effects of the following factors on SoE?

I Height (F1): High vowels > Mid vowels > Low vowels.

I Backness (F2): No significant effect.

I Roundness: Unrounded vowels have lower SoE values (lessstrong voicing) than round vowels.

I These effects are in the opposite direction from theprediction based on the idea that rounding/height decreasevowel aperture.



Summary

Summary: Within Categories

I Compared to tap [R] and stop [d], the trill [r] has the lowestCQH values meaning it patterns more like a fricative.

I Place vs. Sibilance:I Found a significant difference in voicing dep. on place of

articulation but not sibilance.

I Vowels: The expected relationship between vowel height (androundness) and SoE is not borne out; however —

I The reversal in expected SoE values among vowels is alsoobserved among sonorants (nasal > liquids > approx.).



Conclusion & Future Directions

Conclusions

1. Does voicing vary across segment types, and if so;I Yes! Null hypothesis can be rejected.

2. Do these distinctions depend on degree and type ofconstriction?

I Generally speaking, the tighter the constriction → thebreathier the voicing

3. Do these differences reflect distinctions made by(phonological) sonority?

I Only at the broadest level: Vowels > approximants >obstruents

I A number of reversals within each broad manner class.



Acknowledgments

Thank you to Yen Shue for VoiceSauce and Henry Tehrani forEggWorks.

We would also like to thank the students in the UCLA Winter 2015Speech Production graduate course for participating in the designand execution of this study, as well as Soo Jin Park for adding theSoE measure to VoiceSauce.

Thank you also to members of the UCLA Phonetics Lab for helpfulcomments and discussion.



Bickley, C. A. and Stevens, K. (1987). Effects of a vocal tract constriction onthe glottal source: Data from voiced consonants. In Baer, T., Sasaki, C.,and Harris, K., editors, Laryngeal Function in Phonation and Respiration,pages 239–253. College-Hill Press, Boston, MA.

Eduard, S. (1881). Grundzuge der Phonetik. Breitkopf und Hartel, Leipzig.

Fant, G. (1997). The voice source in connected speech. SpeechCommunication, 22:125–139.

Fletcher, M. (1972). Speech and Hearing in Communication. Krieger, RobertE., Huntington.

Hankamer, J. and Aissen, J. (1974). The sonority hierarchy. pages 131–145,Chicago: Chicago Linguistic Society.

Heffner, R.-M. S. (1950). General Phonetics. The University of WisconsinPress, Madison.

Hothorn, T., Hornik, K., and Zeileis, A. (2006). Unbiased recursivepartitioning: A conditional inference framework. Journal of Computationaland Graphical Statistics, (3):651–674.

Jany, C., Gordon, M., Nash, C. M., and Takara, N. (2007). How universal is thesonority hierarchy?: A cross-linguistic study. Paper presented at ICPhS XVI.



Miller, B. (2012). Sonority and the larynx. In Parker, S., editor, The SonorityControversy, pages 257–288. Walter de Gruyter.

Mittal, V. K., Yegnanarayana, B., and Bhaskararao, P. (2014). Study of theeffects of vocal tract constriction on glottal vibration. Journal of theAcoustical Society of America, 136(4):1932–1941.

Murty, K. S. R. and Yegnanarayana, B. (2008). Epoch extraction from speechsignals. IEEE Transactions on Audio, Speech, and Language Processing,16(8):1602–1613.

Nathan, G. S. (1989). Preliminaries to a theory of phonological substance: Thesubstance of sonority. In Corrigan, R., Noonan, M., and Eckman, F., editors,Linguistics categorization, volume 61 of Current Issues in Linguistic Theory,pages 55–68. John Benjamins, Amsterdam.

Ohala, J. J. (1983). The Origin of Sound Patterns in Vocal Tract Constraints.

Parker, S. (2002). Quantifying the Sonority Hierarchy. PhD thesis, Universityof Massachusetts Amherst.

Shue, Y. L., Keating, P. A., Vicenik, C., and Yu, K. (2011). Voicesauce: Aprogram for voice analysis. In Proceedings ICPhS 17, pages 1846–1849.

Sole, M.-J. (2015). Acoustic evidence of articulatory adjustments to sustainvoicing during voiced stops. In Proceedings of the 18th InternationalCongress of Phonetic Sciences. Glasgow, UK: the University of Glasgow.



Stevens, K. (2000). Acoustic Phonetics. Current Studies in Linguistics Series.MIT Press.

Tehrani, H. (2012). Eggworks.http://www.linguistics.ucla.edu/faciliti/sales/software.htm.

Westbury, J. and Keating, P. (1986). On the naturalness of stop consonantvoicing. Journal of Linguistics, 22:145–166.


http://www.linguistics.ucla.edu/faciliti/sales/software.htm

Documents

The relationship between sonority and glottal vibration - …linguistics.ucla.edu/people/keating/Risdal-etal_CUNY2016_final.pdfThe relationship between sonority and glottal ... Introduction