Dept. for Speech, Music and Hearing

Quarterly Progress andStatus Report

Articulatory differencesbetween spoken and sung

vovels in singersSundberg, J.

journal: STL-QPSRvolume: 10number: 1year: 1969pages: 033-046

http://www.speech.kth.se/qpsr

STL-QPSR 1/1969 34.

in the formant frequencies between the spoken and sung vowels were found

to b e of the same type a s irr the previous investigation.(') Thus, the sung

vowels a s compared to the spoken vowels showed the following charact-

er is t ics : a ) the second formant frequency, F2, is lowered in a l l vowels

except in the back vowels [ul, [o] , [a], and [E] ; b) F3 is raised in the

back vowels and lowered in the other vowels; c) F 4 i s lowered in al l

vowels; d) the frequency distance between F3 and F 4 i s reduced in a l l

vowels.

One of the subjects, B4, did not produce these differences a s c lear ly

a s in the previous investigation. Also his articulation differences between

the spoken and sung vowels were much sma l l e r than fo r the other subjects.

A formant frequency analysis was therefore made of vowels produced by

this subject in some Swedish songs with piano accompaniment. Here, the

formant frequencies differed clear ly f rom those of the spoken vowels and

in the same manner a s for the other subjects. Thus, the articulation of

the sung vowels that this subject produced when the X-ray pictures were

taken was not fully representative of his articulation i n actual singing.

Area functions

A study of the acoustical implications of articulatory differences. r e - -

quires knowledge of the a r e a function of the vowel. Therefore, a r e a

functions were worked out fo r the sung and spoken vowels of one of the

subjects, (Bl). The procedure used for deriving a r e a functions f rom the

measures available on l a t e ra l X-ray pictures of the vocal t r ac t is de-

scr ibed in the Appendix.

The agreement between the formant frequencies of the subject and

those measured on the analogue provides an indication that the procedure

fo r deriving a r e a fl~nctions and computing formant frequencies is reason-

ably accurate and essentially correct . Moreover, the resul ts support the

assumption that the vowels were articulated without nasality a l so in

s ing ing . (2)

The effects of articulatory differences were analyzed in the following

way. On the a r e a functions, simulated on the analogue, the a r e a was

reduced by 26 30 over a length of 1.5 cm. The effect of this modification

was systematically evaluated a t ten different places along the vocal tract.

Thus, the f i r s t four formant frequencies were measured af te r each per -

turbation. The percentual changes in formant frequency plotted a s a

STL-QPSR 1/1969 35.

function of the length of the vocal t r ac t inform us about the acoustical sen-

sivity to sma l l ar t iculatory per tuhat ions a t different places in the vocal

tract. The mathematical expression of sac h perturbation s ensivity curves

i s ( 3

where F i s the formant frequency

A F is the formant frequency difference due to the perturbation

A(x) i s the a r e a function

AA(x) is the a r e a difference between the unperturbed and perturbed a r e a function

k is the length of the vocal t r ac t

x is the length coordinate of the perturbation

~ ( x ) i s the volume velocity a t the length l = x

P(x) is the sound p res su re a t the length R=x

p is the density of a i r

c is the sound velocity 1

Experiments with a r e a functions obtained f rom other subjects than B l

showed that, between subjects, the perturbation sensitivity curves differ

only slightly, provided that the place of maximum constriction is located

at the same place.

Perturbation sensitivity curves were worked out fo r the a r e a functions

pertaining to the spoken vowels of subject B 1. In o r d e r to obtain a general

view of the effects of perturbing the vocal t r a c t a t different places, the

perturbation sensitivity curves for each formant of a l l nine vowels were

plotted on the same graph. To make such plots feasible the length was

normalized f rom k = 0 a t the lips and Q = l a t the glottis. These plots,

Fig. 111-A-2a-d, yield the information needed for a study of the acoustical

implications of differences in the articulation of a given vowel.

Fig. 111-A-2d shows, that the subject has to increase his mouth cavity

o r dec rease the volume in the lower pharynx region, if he wants to r i s e

his f i r s t formant frequency. According to Fig. 111-A-2c a narrowing in

the middle of the pharynx will make F 2 to sink substantially in a l l vowels,

Fig . 111-A-2. (a-d) Percen tua l changes i n the f i r s t four fo rmant f requencies due t o a per turbat ion i n s e r t e d a t t e n di f ferent p laces along the vocal t r a c t s imulated on the e lec t r i ca l l ine analogue LEA. The per turbat ion involved a narrowing 6f the a r e a by 26 % over a length of 1 . 5 c m . The a r e a functipns of the Swedish vowels [u] , I

1 [o] , [ a ] , [el, [ e l , [ i ] , [y], [a], and [ b ] give values within I the dashed l ines. I I

AEi .loo F L a

[ul + 8 [ol

+ 6 e [ i l

+ 4

+ 2 0

- 2 - 4

- 6

-8

-10 -12 -Q Norm.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

g 100 4

+6 - +L - + 2 -

0 -2 - -4 -

- 6

A- / %

x\ -L * ,! f'-', - - -

/ 7 '<2" - 7 -.A'

I I I I I I I I I , -1 n 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0

Norm.

*.loo

A +6- +L-

+2 - 0

- 2 -

-6-

- 8

./- / a x ' ~ --

L J - ~ c \ ,d-=

/ \ "'~/- >I-

/ \ -e> f

I I I I I I I I I

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 - ~ 'Nor rn .

STL-QPSR l / l969 36.

but a narrowing of the cross-sect ional a r e a in the mouth will give a r i s e

to F 2 only i f the a r e a is not constricted before a s i s the c a s e in [i]. Fig.

III-A-2b shows, that F 3 can easily be al tered by articulation differences

i n the mouth: a reduced volume immediately behind the incisors and an

expanded volume i n the soft palate region will r i s e F3. Finally, Fig,

III- A-2a i l lustrates , that F 4 is practically uns ensitive to articulation dif-

ferences outside the larynx tube. An expansion of the sinus morgagni and

a narrowing of the entrance to the larynx tube lowers F 4 essentially. We

will next proceed to study how the subject uses these possibilities to a l t e r

his formant frequencies in singing.

Articulatory differences

F r o m a physiological point of view there a r e severa l elements in the

ar t iculatory organs that a r e independently variable. In some of these

dimensions, notably (1) larynx position, (2) the jaw and the l ip openings,

(3) the corss-sect ional a r e a of the vocal t r ac t in the velum region, and (4)

the tongue shape in the back vowels, typical differences were found be-

tween spoken and sung vowels. We will now descr ibe these ar t iculatory

differences and discuss the i r consequences a s regards the formant f re -

quencies,

(1) The position of the sagittal mid point of the r ima glottis related to the

lower frontal co rne r of the second cervical ver tebra is shown in Fig.

III-A-3a. In speech the larynx assumes a high position in [ e l and [i]

whereas the rounded front vowels C y ] , [u], and [b] a r e articulated with

a lower larynx. 4, high larynx position in vowels where the tongue is

pulled upwards seems natural, since the tongue root and the larynx a r e

attached to each other via the hyoid bone and the epiglottis. Our acoustic

analysis indicates that larynx depression would be specially effective in

C y ] . In singing, however, the big differences in larynx positions a r e

smoothed out and the average position is lower ( ~ i g . 111-A-3b). This is in

accordance with the findings of Faaborg-Andersen & Vennard, that muscles

lowering the larynx a r e active in singing, (4)

Singing teachers often recommend the i r pupils to sing [i] a s [y].

This might be a simple way to make the pupil lower his larynx, since the

normal speech position of the larynx in L y ] is much lower than in [i].

PL lowering of the larynx implies a lengthening of the pharynx par t of

the vocal t ract . The X-ray pictures showed that this length expansion

Fig. 111-A-3. (a) Larynx position i n different Swedish vowels. Solid lines: spoken vowels, dashed l ines: sung vowels. Subjects: B1, n B3, A B4.

(b) Difference i n larynx position between spoken and sung Swedish vowels. Subjects: *B1, X B3, A R4. The solid line shows the mean values.

(c) Changes i n the formant frequencies due to a 1 c m lowering of the larynx a s measured on the e lec t r i ca l line analogue LEA.

STL-QPSR 1/1969

takes place immediately above the epiglottis. A lowering of the larynx

by 1 c m could therefore be simulated by simply inserting two extra length

sections at the corresponding place on the analogue. The changes in the

formant frequencies obtained a r e shown on Fig. 111-A-3c. F 3 is the leas t

sensit ive measured formant i r respect ive of vowel. F 1 is the most sen-

sit ive formant in the back vowels, and F 2 is the most sensit ive in the front

vowels. F 4 is r a the r sensitive for a l l vowels.

Thus, the lowered F2 in the sung vowels may be an effect of the low-

ered larynx, since the degree of larynx lowering a s well a s the sensitivity

of this formant to a lowered larynx both culminate in the front vowel ser ies .

Also the reduced frequency distance in the sung vowels between F 3 and F4

may be partly explained with reference to the larynx lowering, since F4 i s

much m o r e sensit ive than F 3 in a l l vowels.

(2) The l ip and the jaw openings were examined not only on the X-ray pic-

tures , but a l so i n a s e r i e s of frontal and la te ra l photos, a t leas t two for

every spoken and sung vowel. F r o m these photos i t is possible to deduce

the relations between the width of the mouth opening and its height and the

position of the mouth corners . These relations a r e required fo r the

estimation of the l ip opening area.

The lip opening seems to be strongly dependent on the vocal effort.

This c ircumstance appear likely when we examine Fig. 111-A-4 which

shows a comparison of the degree of jaw opening a t various efforts. (6)

As effort increases , jaw opening - and thus, by implication, l ip opening - a lso tends to increase. The depression of the mandible was observed to

be l a r g e r in the sung vowels. The sound levels of the spoken vowels were

lower than those of the sung vowels. The vocal effort was therefore pre-

sumably bigger when the subjects sung than when they spoke the vowels.

Thus, the l a r g e r l ip opening in the sung vowels, evident on Fig. 111-A-5a,

and the lowered jaw could be effects of a difference in the vocal effort

ra ther than charac ter i s t ics of the sung vowels.

If, however, the l ip opening in the sung vowels tends to be of the same

s i ze a s in the spoken vowels o r smal le r , in spite of an increased jaw open-

ing, a s is the c a s e with [el and [i] , this is likely to be a significant

feature of the articulation of these vowels in singing. Another character is t -

i c of the vowels [ e l and [il in singing is a considerably increased protru-

sion of the lips. This effect is a l so observed in L y ] but is negligible in the

DEPRESSION OF MANDIBLE

(strong and weak e f f o r t )

-. -. 'A. /.

%..J 1.

Fig . 111-A-5. (a ) Ratio between the l ip opening a r e a of spoken (A ) and sung (A ) vowels. Subjects:

Su B1, B3, s P B4. The solid line shows the mean values.

(b) The difference in mouth c o r n e r position between spoken and sung vowels. Subjects: B1, B3, B4. The solid line shows the mean values.

( c ) Changes in the fo rmant frequencies due to a n i nc r ea se in L / A ~ by 50 70 as measured on the e lec t r i c line analogue LER.

STL-QPSR 1/1969 3 9.

The velum is situated a t about .35&-. 45R in the normalized vocal t ract .

The acoustic consequences of an a r e a expansion in this region a r e i l lus-

t ra ted in Fig. 111-A-2a-d. F 1 will r i s e in the front vowels and be ra ther

unaffected in the back vowels. F Z will d rop a l i t t le in a l l vowels. F 3 will

tend to r i se , especially in the back vowels. Thus, the lowering of the

larynx and the a r e a expansion in the velum region both contribute to the

lowering of F 2 in the sung vowels. The r i s e in F 3 of the back vowels may

to a cer ta in extent b e a n effect of the same articulatory difference,

(4) The tongue shape is changed considerably only in the back vowels. In

[u] and [o] the tongue tip i s given a m o r e frontal position i n singing than

in speech. At the same t ime the tongue root is pulled downwards by the

slightly lowered larynx. The velar tongue shape in [u] and [o] is thus

replaced by a pharyngeal shape, more resembling the tongue shape of an

[a]. This means, that the ent i re fo rm of the frontal cavity is al tered in

the way demonstrated in Fig. 111-A-1. F o r the spoken back vowels the

la rges t a r e a is found immediately behind the incisors , and for the sung

vowels the maximal values a r e situated in the ve lar region of the mouth,

s ince the tongue tip fills out a good dea l of the frontal par t of it. These

changes mainly affect F3 , a s is seen on Fig. 111-A-Za-d. The filled out

frontal par t of the mouth cavity and the expanded poster ior par t of it both

tend to r i s e F3. The effect on other formant frequencies i s small. Here

we have a probable explanation of the raised F 3 of the sung back vowels.

Discussion

Above we have found support for ascribing cer ta in formant frequency

differences between spoken and sung vowels to cer ta in articulatory dif-

ferences: a ) The lower F 2 in the front vowels i s an effect of the lowered

larynx and i t s physiological consequence in the velum region. b) The

r i s e in F 3 in the back vowels is an effect of an al tered tongue shape in-

volving a decrease of the cavity immediately behind the incisors and an

increase of the poster ior pa r t of the mouth cavity. In the front vowels

the lowered F 3 is an effect of the lowered larynx and, a s regards the

spread front vowels, a l so of an increased lip protrusion. c) The lowered

F4 is an effect of the lowered larynx. d) The reduced frequency distance

between F3 and F4 is a consequence of ar t iculatory differences mentioned

under d ) and c).

STL-QPSR 1/1969 40.

Let us now discuss why the spoken and sung vowels a r e articulated dif-

ferently. I s it in o rde r to achieve cer tain esthetical, o r bet ter , acoustical

charac ter i s t ics of the sung note, o r is i t fo r physiological reasons?

It is a common observation that the sound level in the frequency range

2.5-3 kHz is abnormally high in sung vowels. ( I1) This increase is often

re fer red to a s the "singing formant". When F 2 is low in frequency, a s in

the back vowels, the spectral level in this frequency range i s low. (12)

However, i t can be raised i f the higher formants a r e brought c lose r in

frequency. This is i l lustrated in Fig. 111-A-6a, showing the spec t ra l en-

velope of an [u] with the formant values normal fo r this vowel according

to Fant, and with the source character is t ic -10 d ~ / o c t . ( I 3 ) The graph

a lso shows what happens to the spec t ra l level if F3, F4, and F 5 a r e brought

c loser in frequency. It i s seen on Fig. 111-A-6b that this spectral enve-

lope matches an actually sung vowel spec t rum envelope of an [u]. Experi-

ments showed that the "singing formant" can be explained in the same man-

n e r in other vowels.

The "singing formant" appears especially i n the back vowels to give

the t imbre "brilliance". The s inger seems to bring about this spectrum

character is t ic by a skilled control of the higher formant complex. He

does not nasalize and the source spec t rum appears to be normal. How-

ever , needless to say, other s ingers may employ different methods.

So f a r , we have found a plausible reason for the articulatory modifica-

tions of the back vowels. In the front vowels, the big difference appears

i n the larynx position and in the l ip opening. Probably, the larynx lowering

i s made for physiological reasons. It is well known that singing with a

high larynx position may damage the vocal cords. (14) A reason fo r this

might be that a low larynx position makes the t issues between the thy~2ic!

and cricoid car t i lages m o r e lax, The shifts in the relative position of

these two car t i lages, required for the big pitch variations in singing,

could be made with l e s s muscular activity, if the t i ssues joining them a r e

lax. ( I5 ) However, this is a question for future research. The differences

i n the l ip opening might be due to habits associated with the adjustment of

the larynx position.

Finally, i t shal l be pointed out, that this investigation i s based on ob-

servations f rom a restr ic ted number of bass s ingers singing a r a the r low

note. It is of course possible that the observed articulation i n singing is

dependent on the type of voice, the pitch, and the shape of the vocal tract.

__*

kHz

Fig . 111-A-6. ( a ) Spec t ra l envelopes of the vowel [ u ] s imulated on a t e r m i n a l analogue and demonstra t ing the effects of a r e - duced frequency dis tance between the th i rd , fourth, and fifth fo rmants . Source charac te r i s t i c : -10 d ~ / o c t .

(b) Spec t rum of a sung [u]. The dashed c u r v e i s the s a m e as the chain dashed c u r v e on the upper f igure.

STL-QPSR 1/1969 41.

Acknowledgments

This work was made possible by support f rom the Tri-Centennial

Fund of the Bank of Sweden and by valuable discussions with Prof. G.

Fant and Dr. B. Lindblom a t the Dept. of Speech Communication, KTH.

Dr. K. Lagergren and his staff a t the Radiological Dept. D a t Karolinska

Sjukhuset, Stockholm, and Mr. G. Garpendahl of the Dept. of Speech

Communication, KTH, a r e gratefully acknowledged fo r their kind a s -

s is tance in taking the X-ray pictures vin, the l ip photos. The author

a l s o wishes to thank the s ingers fo r the i r kind cooperation.

References:

(1) Sundber J . : "Formant frequencies of bass singers", STL-QPSR 17i968, pp. 1-6.

(2) Vennard, W. : "An experiment to evaluate the importance of nasal resonance in singing", Folia Phoniat. - 16 (1964)~ p. 146 ff.

(3) Fant , G.: "Kompendium i Taloverfiiring", ( ~ o t e s for a course in speech communication), Dept. of Speech Communication, KTH, 1967 (2nd edition 1969); s ee a l so Schroeder, M. : "Acoustic measurements of vocal-tract shapes", J. Acoust. Soc. Am. 41 (1967)~ p. 1002 ff. o r Heinz, J.M.: "Perturbation funct ions fo r the determination of vocal-tract a r e a functions f rom vocal- t r ac t eigenvalues", STL-QPSR 1/1967, pp. 1 - 14.

(4) Faaborg-Andersen, K and Vennard, W.: "Electromyography of extrinsic laryngeal muscles during phonation of different vowels", Ann. of Otol., Rhinol. and Laryngol. - 73 (1964), p. 248 ff.

(5) Fromkin, V. : "Paramete r s of lip-pdsitizn", Working Paper s i n Phonetics, UCLA, June 1964.

(6) Personal communication with Dr. B. Lindblom, who a lso offered Fig. III-A-4a.

(7) Perkell , J. S. : "Physiology of Speech Production: Results and Implications of a Guantitative Cineradiographic Study", The Massachusetts Institute of Technology (1 968).

(8) Fant , G. : Acoustic Theory of Speech Production ( ' s-Gravenhage 1960).

(9) Lindblom, B. and Sundberg, J. : I f A quantitative model of vowel pro- duction and the distinctive features of Swedish vowelsf1, STL- QPSR 1/1969, pp. 14-32.

(10) Sovijarvi, A. : "Die gehaltenen, gefliisterten und gesungenen Vokale und Nasale d e r finnischen Sprache", Annals Academiae Scientiarum Fennicae B XLIV, 2, Helsinki 1938.

(1 1) See, for instance, Vennard, W. : Singing, the Mechanism and the Technique (New York 1967).

(12) Fant, G. , Fintoft, K., Liljencrants, J . , Lindblom, B., and Mdrtony, J. : "Formant-amplitude measurements:' J. Acoust. Soc.Am. 35 (1963)~ p. 1753 ff. -

(13) Fant, G,: "Acoustic analysis and synthesis of speech with applica- tions to Swedish", Er icsson Technics No. 1, 1959,

(14) Luchsinger, R. and Arnold, G. E. : Lehrbuch d e r Stimm- und Sprachheilkunde (Wien 1959).

(15) Suggested by Prof. G. Fant, personal communication (1969).

APPENDIX

On the problem of obtaining area functions from lateral X-ray

pictures of the vocal t ract

The problem of deriving an area function from the measures available

on lateral X-ray pictures of the human vocal t ract has been dealt with by

several authors. 2' 3, A method for obtaining measures from X-ray

pictures i s described by Heinz & Stevens and ahman (3' 4). A polar co-

ordinate system i s used the center of which i s the center of a circle I

tangent to the horizontal portion of the hard palate and the back pharynx

wall.

It has been stated, that the cross-sectional a rea in the mouth region,

A ~ ' i s a power function of the distance a between the hard palate and M the tongue.

where k and a a r e constants. According to Heinz & Stevens, these con-

stants have different values in different parts of the mouth. The constants

were determined frorn plaster casts of the mouth by measuring the a reas

for different positions of the tongue.

Such measurements have recently been made by the author on plaster

casts of three different subjects' mouths in order to determine the varia-

tions of the constants. These constants were found to have about the

same values regardless of the place in the mouth, as i s evident from the

graphs on Fig. la-c. Fig. l b shows the function obtained from the de-

scribed polar coordinate system as well as measures obtained when the

areas were measured at an angle normal to the upper teeth contour. The

figure indicates that only small e r ro r s a r e made, i f the constants k and b

a r e assumed to have the same values in the entire mouth region, and that

the angle to the mid line of the vocal t ract i s of minor importance.

Heinz & Stevens obtained good results estimating the cross-sectional

areas in the pharynx region, Aph, as the area of an ellipse. The dis-

tance between tongue contour and back pharynx wall, a i s taken a s Ph' one axis and the other axis varies a s follows: 3 c m in the upper pharynx

region, 2.5 cm behind the epiglottis, 2.0 cm between the epiglottis and

the entrance to the larynx tube, and 1.5 cm in the larynx tube,

DISTANCE BACK PHARYNX WALL- TONGUE

Fig. 2 (Appendix). (a ) The major axis of the equivalent ell ipse giving the c ros s -sectional a r e a in the pharynx region, if the anter ior-poster ior measure i s used a s the minor axis . The major axis i s plotted a s a function of the back pharynx wall-tongue distance.

(b) The cross-sec t iona l a r e a in the pharynx region plotted a s a function of the back pharynx wall-tongue distance.

STL-QPSR 1/1969

( ~ ~ p e n d i x )

of the subject. Fur thermore , the quantization s teps of the analogue were

too big to achieve the required accuracy of the a rea . The a r e a function of

the larynx tube was therefore given a f o r m that gave a co r rec t value of

the fourth formant frequency,

The differences between the formant frequencies of the subject and

those measured on the analogue a r e given in Table I. The differences

a r e ra ther sma l l in spite of the ra ther crude method.

Big percentual differences appear only fo r the first formant, especially

fo r the front vowels. These negative differences might be due to the for -

mant f r e uency raising effect caused by the wall inductances a s discussed

by Fant. 1 5 )

References:

(1) Chiba, T, and Kajiyama, M. : The Vowel, i t s Nature and Structure ( ~ o k y o 1941).

(2) Fant, G. : Acoustic Theory of Speech Production (' s-Gravenhage 1960).

(3 ) Heinz, J. M. and Stevens, K. N. : "On the derivation of a r e a functions and acoustic spec t ra f rom cine radiographic fi lms of speech", P a p e r S4 presented at a meeting of the Acoustical Society of America i n May 1964.

(4) ahman, S . : "Numerical model of coarticulation", J. Acoust. Soc. Am. - 41 (1967)~ p. 310 ff.

(5) Fant, G. : "Formants and cavities", Proc . of the Fifth International Congress of Phonetic Sciences, Miinster 1964, pp. 120,141 (Base1 1965).

(6) Fromkin, V. : "Parameters of l ip position", Working Paper s in Phonetics, UCLA, June 1964.

STL-QPSR 1/196 9

TABLE I

Vowel

F1

Sub- Syn- Differ. ject thesis ence

Hz Hz 70

F2

Sub- Syn.. Differ- ject thesis ence

The formant frequencies of the subject B1 and those of the derived

area functions, measured on the electrical line analogue LEA. Sp

refers to spoken vowels and Si to the sung vowels.

F3

Sub- Syn- Differ- ject thesis ence

kHz kHz yo kHz kHz %