Delattre, Pierre. Studies in French and Comparative Phonetics – T28

The Physiological Interpretation of
Sound Spectrograms *1

In the physiological study of speech articulations our most objective information has
come, until recently, from radiograms. 12 Now we have, in addition, spectrograms, 23
which, if we learn to interpret them, can also give us very objective information. For
the typical phase (portion) of a speech sound, the interest of a spectrogram may be
about equal to that of a radiogram; but for the transitional phases, the interest of a
spectrogram will probably be much superior; and from the practical viewpoint of
availability, the spectrogram will have a marked advantage for it can be had in a few
minutes and at low cost. But to the linguist, the usefulness of a spectrogram depends
on his ability to interpret it in articulatory terms. We need not stress, therefore, the
importance of investigating the relation between formant 34 positions and speech organ
positions at this stage of the still young science of sound spectrography.

Since Martin Joos's presentation of the relation between the articulatory triangle
and the acoustic triangle in terms of a relation between formant 1 and tongue-height
and between formant 2 and back-to-front tongue placement, 45 some progress must
have been made by spectrography researchers. This is our contribution to this
progress.

It is generally believed that “the shape of the filtering cavity is so very complex as
225to be mathematically unmanageable …” 56 This may seem very discouraging, but it is
not. To the phonetician it means only that it will probably not be possible to determine
exactly to what extent a certain formant can be assigned to a certain cavity.
This is not of great importance to him. What he needs to know is rather the relation
of each formant to the position or movement of articulatory organs. And that can
be determined to a large extent (a) by studying the effect of isolated articulatory
movements on formant positions with the help of spectrograms, (b) by comparing
the spectrograms with radiograms, and in certain cases (c) by checking the findings
on a speech synthesizer to see if the formant changes (or change) resulting from an
isolated articulatory movement produced the auditory impression expected from this
movement.

We shall apply this (a), (b), (c) technique successively to formants 1, 2, and 3 in
that order. Before starting, let it be well understood that we are not concerned here
with mouth cavities; to formant frequencies, we are exclusively trying to relate
articulatory movements and positions. (More will be said about this in the discussion
of formant 2).

Formant 1

The phonetic triangles and quadrilaterals, in their vertical direction, have all been
based on tongue-height (the highest point of the tongue arch); therefore, it was
natural for Joos to speak of tongue-height when relating the triangle obtained by
plotting formants 1 and 2 on a logarithmic scale to the traditional phonetic (articulatory)
triangle. However, the term tongue-height may not be the most appropriate in
relation to the frequency of formant 1. Perhaps the more general term of “opening”
(meaning overall opening of the oral tract with definite relation to the width of the
strictures at the main points of articulation, but not depending entirely on that)
would be more correct. If we examine formant 1 not only on spectrums of vowels
but also on those of voiced consonants, this will appear clearly.

Vowels.

This similarity of [i] and [u] formant 1 frequencies does not correspond to tongue-height
as known through radiography. The articulatory quadrilaterals of Jones 78 and
Kenyon 89 both give [u] a tongue-height much lower than that of [i] and very nearly as
low as that of [e] (Fig. 2, a and b). And this tongue-height difference between [i] and
[u] is not even as marked as in actual sets of x-rays that we have consulted. Parmenter's
radiograms of American vowels 910 show totally different tongue-heights for [i]
227and [u]: if we measure the distance in mm. from the highest point of the tongue to
the highest point of the palate, we rind that, with [a] at 8.5 and [e] at 3.5, [i] is at 1.5
and [u] at 5.0 (Fig. 2c). On the Holbrook-Carmody x-rays of superimposed vowels
from many languages, 1011 the same measurements (on another scale) give: with [a] at
24 and [e] et 10, [i] at 5 and [u] at 13 (Fig. 2d). On the Czech vowels of M. Hala, as
presented by Adrien Millet, 1112 similar measurements on a different scale give: with [a]
at 12 and [e] at 8, [i] at 3 and [u] at 8 (Fig. 2e). On the French vowels x-rayed by
Oscar G. Russell, 1213 with [a] at 9 and [e] at 4, [i] is at 2 and [u] at 5 (Fig. 2f). All these
radiograms concord rather closely in presenting a considerable difference in tongue-height
between [i] and [u] — even more difference than on the Jones quadrilateral.

If we compare two other vowels, such as [e] and [o], in the same manner, we find
almost the same aspect. Briefly, [e] and [o] have very nearly the same frequencies for
formant 1 on the spectrograms, and their synthetic production is done satisfactorily
by using the same formant 1 frequency for both; but their tongue-heights show
differences about comparable to those between [i] and [u]. For instance, on the
Parmenter x-rays, with [a] at 8.5, [e] is at 3.5 and [o] at 7.0. On the Holbrook-Carmody
x-rays, with [a] at 24, [e] is at 10 and [o] at 19.

All this confirms that tongue-height differences and frequency differences of formants
1 for different vowels do not correspond when comparing articulatory triangles
with acoustic triangles. If we look for some feature that is nearly the same for [i] and
[u], for [e] and [o], in the articulation, we find it better in the overall opening of the
mouth tract, as measured for instance by the distances between the upper and lower
incisors or by the distances between the highest point of the tongue and the point of
the palate closest to it — in other words, the general width of the structures. Either of
these two measurements show only small differences between corresponding front
and back vowels. The use of the term opening instead of tongue-height seems indicated.

Consonants.

Consonants. The voice bar — or formant 1 — of voiced consonants serves to further
confirm that tongue-height is not the most appropriate term in relation to formant 1
frequency. First we notice, by comparisons with one or another of the consonants
that have the most clearly delimited voice bars, that the frequency of formant 1 is in
accord with the accepted notion of degrees in opening of the mouth: judging from a
large set of consonant spectrograms we made at the Bell Telephone Laboratories of
New York (summer of 1947) (Fig. 3), the frequency of formant 1 for [b] and [d], is
always lower than for [m] and [n], and the latter is always lower than for [l].

Conclusion for formant 1

Conclusion for formant 1. The relation between formant 1 position and articulatory
position should be stated in the following terms: There is a direct relation between
formant 1 frequency rising and overall opening of the oral tract. The higher the
frequency of formant 1, the wider the overall opening; and inversely.

Formant 2

In those two remarks lie the main causes of discrepancy between the traditional
vowel charts (articulatory) and the acoustic vowel charts such as that of Fig. 1.

Joos, with the traditional tongue placement triangle in mind, has already made clear
the general relation that exists between formant 2 and back-to-front tongue placement.
We wish 1) to further discuss this relation of formant 2 with back-to-front
tongue placement; 2) to establish another striking relation of formant 2, the one it has
with lip spreading-to-rounding; 3) to discuss the conjugating of these two notions
into a single one.

1. We shall see here that there is a direct relation between tongue backing and
formant 2 frequency lowering, but not quite in the same sense as is implied by the
front-to-back horizontal direction of the phonetic triangle or quadrilateral. In relation
to formant 2 lowering, tongue backing is not measured by how far back the highest
point of the tongue is (as in the phonetic triangle) but by how far back-and-up the
tongue as a whole is retracted. This retraction cannot be measured as well on radiograms
as the highest point of the tongue but it does show clearly (as we shall
demonstrate farther on), and besides it is felt kinesthetically much better than the
highest point of the tongue (one feels that the tongue is pulled back-and-up more
for [u] than for [ɔ]).229

To exemlipfy this we should compare pairs of vowels that differ by tongue backing
only, all other conditions remaining practically equal. Such are the three French pairs:
[y]-[u], [ø]-[o], [œ]-[ɔ] (Fig. 4). From [y] to [u] (Fig. 4a), the lips remain equally
rounded and the jaws equally closed; the only important change is in the tongue
position which passes from extreme front-and-up to extreme back-and-up. On the
spectrogram this backing of the tongue is translated by a marked lowering of formant
2 while formant 1 remains in the same place. Exactly the same could be said of the
transitions from [ø] to [o] (Fig. 4b) and from [œ] to [ɔ] (Fig. 4c).

Comparisons can also be made satisfactorily regardless of the degree of opening as
long as the vowels agree from the angle of rounding-spreading. In the series [i], [e], [ɛ]
(Fig. 1, top left), where all the vowels have lip spreading, formant 2 lowering is in
accord with tongue backing. In the French series [y], [ø], [œ] (Fig. 1, top center) and
[ɔ], [o], [u] (Fig. 1 top right), where all the vowels have definite lip rounding, formant 2
lowering again goes with tongue backing.

For the last three vowels [ɔ], [o], [u], however, the traditional vowel quadrilateral
does not agree with the acoustic feature of formant 2 lowering (compare [ɔ], [o], [u]
on Fig. 2 with [ɔ], [o], [u] on Fig. 1). The Jones quadrilateral shows [u] less far back
than [o], and [o] less far back than [ɔ] (the way the charts are generally disposed, [u]
is left of [o], and [o] left of [ɔ]). There are two striking reasons for this discrepancy
between the Jones chart (Fig. 2) and the acoustic chart (Fig. 1). (a) The Jones chart
does not take into account the important feature of rounding, which, as we shall see
next, also has a lowering effect on formant 2. For instance, if [u], [o], [ɔ] had the same
tongue backing (as is the case on simplified quadrilateral), on the acoustic chart,
[o] would still be on the right of [ɔ] because is it more rounded, and [u] would still
be on the right of [o] for the same reason. (b) The Jones chart bases its notion of
tongue backing on the backing of the highest point of the tongue arch and not on
actual back-and-up retraction of the tongue as it is felt kinesthetically.

2. Let us examine the relation between lip rounding and formant 2 lowering. To
study it we may take any pair of vowels in which the two sounds differ by lip rounding
only and are about similar otherwise. Fig. 4 offers us 3 such pairs, [i] and [y] have
about same opening and same tongue fronting, but [i] has spread lips and [y] rounded
lips: passing from [i] to [y], a clear lowering of formant 2 can be observed on the
spectrogram (Fig. 4a). The same can be said of the pairs: [e]-[ø] (Fig. 4b), and
[ɛ]-[œ] (Fig. 4c).

We may conclude that there is a direct relation between lip rounding and frequency
lowering of formant 2: as the lips are rounding formant 2 is lowering, and inversely.

3. Trying to find a common denominator for the two preceding relations (tongue
backing and lip rounding) to formant 2 lowering, we noticed that both tongue
backing and lip rounding had the effect of lengthening the front (or mouth) cavity
(Fig. 4a, b, c): the highest frequency for formant 2 is obtained for vowel [i] which
seems to have the shortest possible front cavity (with maximum tongue fronting and
maximum lip spreading); the lowest frequency for formant 2 is obtained for vowel [u]
which seems to have the longest possible front cavity (with maximum tongue backing
and maximum lip spreading). We stated this at the MLA meeting of 1947 (in our
paper on the nasal resonances of French nasal vowels) and again in an article that was
to serve as a simple introduction to spectrography for French teachers. “Il existe une
relation constante et inverse entre la hauteur de la formante 2 et la longueur de la cavité
de résonance buccale.” 1314 To put this in terms similar to those that we have been using
here we should say: there exists a direct relation between formant 2 lowering and front
cavity lengthening. This is a good statement only if “cavity lengthening” is interpreted
appropriately. Cavity lengthening is not an acoustic feature here but a physiological
one: using the terminology “cavity lengthening” permits to include two physiological
movements (tongue backing and lip rounding) into one statement. In fact it must
mean tongue backing and/or lip rounding. It does not imply any notion as to the size
of the cavity. For instance, should it imply that by lengthening the cavity becomes
larger it would also imply that formant 2 is independently related to the size of the
front cavity. And that is not true or at least is not known to be true: because the speech
cavities are “mathematically unmanageable” we don't know to what extent formant 2
231is related to the front cavity; we only know that theoretically any formant is related to
the whole System of speech cavities. It is probable that the relation between formant 2
and front cavity varies for each vowel. The opinion of H. K. Dunn of the Bell Telephone
Laboratories is that the farther apart formants 1 and 2 are the more they can be
assigned respectively to the back and the front cavities, and the nearer they are to one
another the more they both must be assigned equally to both cavities. This is about
the limit of what can be known at present of the degree of independent relations
between cavities and formants.

Formant 3

Let us examine now two cases of simple relation between formant 3 and some
articulatory movement.

1. Velum relation to formant 3. (The velum motions are involuntary. We do not
feel them and cannot control them directly. But they obey our seeking to produce
nasality and we can control them indirectly by doing just that. Therefore, “lowering
the velum, ” here, will mean “seeking to nasalize”).

Starting from an oral vowel, if we lower the velum while holding all other speech
organs immobile, the frequency of formant 3 rises considerably while the frequencies
of formants 1 and 2 remain fairly stable. Fig. 6 shows this rise of formant 3 in nasalizing
of an [ɛ]. 1617 (We are speaking only of frequencies, not of the intensities or
modes of those formants, here.) This rise of formant 3 averages very close to 300
cycles in the case of French nasal vowels as compared with oral ones. It is a little
more marked for [ɑ̃] than for the three other nasal vowels [ɛ̃], [œ̃], [ɔ̃].

The opposite experiment confirms the one above. If, for instance, a French nasal
vowel is denasalized by raising the velum while trying to hold all other organs immobile,
the frequency of formant 3 lowers by some 300 cycles while the frequencies
for formants 1 and 2 remain practically stable. Fig. 6b, c, d, e, show this lowering of
formant 3 in the denasalizing of [ɛ̃], [œ̃], [ɔ̃] and [ɑ̃]. 1718

We conclude that there exists a direct relation between the frequency rising of
formant 3 above its normal level and the lowering of the velum as it is lowered in
nasalizing.

We limit our conclusion to the case where the velum is lowered “as in nasalizing, ”
that is, with the back part of the velum away from the wall of the pharynx so as to
allow the nasal cavity to communicate with the oral cavity. It is possible that lowering
the velum without the extremity leaving the wall of the pharynx be also related to the
rising of formant 3, but it is not probable. We experimented with French [ʀ], which
requires the back of the tongue and the velum to draw toward one another. In the case
of the front rounded vowels, the transitions to [ʀ] seemed to indicate, in addition to
tongue backing shown by lowering formant 2, velum lowering shown by raising
formant 3. So, it is very tempting to interpret the inverse sinuosities of formants 2 and
3 as the movements of tongue and velum drawing together. However this interpretation
is not upheld by the examination of transitions of all other vowels to [ʀ].
For instance, the [a] transition to [ʀ] shows no rise of formants 3, and the [i] transition
to [ʀ] even shows a lowering of formant 3.

2. We mentioned above how r-color can be added to [a] by lowering the frequency
of formant 3 on a speech synthesizer such as the Cooper pattern playback. To add
this r-color to any vowel, in human speech, it is sufficient to raise the tip of the tongue
(let it take the well known retroflex position) while making no effort to change the
positions of any other organs. Fig. 7 shows three transitions from an ordinary vowel to
an r-colored vowel by simple raising of the tongue tip. The vowels of Figs. 7a and 7b
were uttered by an American from Michigan. The vowel from Fig. 7c was uttered by a
Frenchman. Spectrums of the tongue raising transition always show a frequency
lowering of formant 3. The range of this frequency change is generally considerable;
it may reach 1000 cycles.

Conclusion

Conclusion: There exists a direct relation between the frequency lowering of
formant 3 below its normal level and the raising of the tongue tip toward a retroflex
position as in the articulation of Midwestern-American r.

However the effect of tongue tip raising is not always limited to formant 3. Formant
2 is also affected in some cases. When formant 2 is already close to formant 3 as for
French [y] and [ø], tongue tip raising lowers formant 2 alongside formant 3. More
generally, when formant 2 is higher than for [ə], tongue raising tends to lower it
toward its [ə] frequency, and when formant 2 is lower than for [ə], tongue tip raising
tends to cause it to rise toward its [ə] frequency.

Formant 1 also seems to be affected by tongue tip raising but it really is not, at
least not directly. When tongue tip is raised, formant 1 tends toward the frequency it
234has for [e], approximately. But this can be overcome by keeping the general opening
of the mouth very stable. Therefore we may say that tongue tip raising affects formant
1 only if mouth closing or opening takes place at the same time.

Summary

We have tried to bring out the articulatory meaning of formants 1, 2, and 3 of sound
spectrograms in a discussion of their relations with tongue-height, mouth opening,
tongue backing, lip rounding, front cavity lengthening, velum lowering, and tongue
tip raising; and to show that direct relations exist between the following formant frequency
changes and the following articulatory movements: 1) between formant 1
frequency raising and overall mouth opening: 2) between formant 2 frequency lowering
and tongue backing; 3) between formant 2 frequency lowering and lip rounding;
4) between formant 2 frequency lowering and front cavity lengthening, this being a
manner of conjugating statements 2 and 3 into a single one (cavity length, here, has a
physiological meaning only, not an acoustical one); 5) between formant 3 frequency
raising and velum lowering as in nasalizing; 6) between formant 3 frequency lowering
and tongue tip raising as in r-coloring.235