Published online by Cambridge University Press: 22 April 2026
This experiment was undertaken to determine whether a definite relation exists in French between the duration of a vowel and its variations in color. Conforming with the terminology of Martin Joos, we use the word color rather than timbre or vowel quality. Within a limited range any vowel sound may undergo some variation in the frequency of its formants without losing its identity. This study will attempt to show that in French a definite relation exists between these frequency variations and the length of an [a] sound.
Some 445 spectrograms of the French sound [a] were examined. The samples were of the broad-band or 300-cycle-filter type—all recorded from pronunciations of a single native informant. At the time the spectrograms were made, the speaker made no conscious effort to preserve the color of the vowel. He was primarily concerned with the pronunciation of the consonants which sometimes preceded and sometimes followed the [a] sounds in question.
1 First reported in a paper read before the 1949 meeting of the Modern Language Association of America at Stanford University.
2 The color of a sound is represented on a spectrogram by three horizontal bands or formants appearing at certain levels one above the other. The heights of these formants represent certain resonance frequencies and are relatively constant for each vowel or consonant sound. A characteristic picture is thereby realized for each sound. For basic acoustic phonetic theory see Martin Joos, Acoustic phonetics (Language Monograph No. 23, 1948).
3 The frequency range of each spectrogram extends from 0 to 3500 cycles. The narrowband method of filtering reproduces the formants at a width of 45 cycles. A better picture of the resonance frequencies is obtained with the broad-band filter, which reproduces each formant at a width of 300 cycles.
4 Pierre Fouché, La prononciation actuelle du français (ou en sont les études de français), Le français moderne 1.59 (1933): ‘Dans le français correct, l'accent ne quitte jamais la dernière syllabe.‘
5 Pierre Delattre, L'accent final en français, The French Review 12.145 (1938): ‘La durée est le seul des trois éléments acoustiques (intensité, hauteur, durée) qui soit toujours, par sa proéminence, un facteur de l'accent.‘
6 For evidence supporting this statement see the intensity graph for French and English, C. E. Parmenter and A. V. Blanc, An experimental study of accent in French and English, PMLA 48.607 (1933).
7 The three formants. Beginning with the one of lowest frequency, the formants are numbered respectively 1, 2, and 3.
8 In converting the statistical polygon into a normal curve, the height or peak of the mean is obtained by the formula
![]()
where Yo is equal to the height of the normal curve at the mean formant frequency, N represents the total number of cases measured, and small sigma denotes the standard deviation in cycles per second. The various other heights of the plus and minus values along the horizontal scale or abscissa are fractions of Yo and may be compared as follows: Y at ±1σ = .60653 × Yo; Y at ±2σ = .13534 × Yo; Y at ±3σ = .01111 × Yo. For a full discussion of this procedure see Herbert Sorenson, Statistics for students of psychology and education 169-70 (New York, 1936).
9 As shown by the formula given in fn. 8, N is directly proportional to Yo. An increase in the total number of cases measured would produce a corresponding increase in the height of the normal curve at the mean.
10 Joos, Acoustic phonetics 57-9.
11 P. Delattre, Un triangle acoustique des voyelles orales du français, The French Review 21.477-84 (1948).
12 Actually Fig. D represents the Delattre acoustic vowel triangle and our average accented and unaccented [a]'s superimposed together on the Joos formant chart.
13 Joos, Acoustic phonetics 101-2, discussing the problem of the pure sample, already observed the striking effect of consonants on the formants of contiguous vowels: [g] shifts the vowel [1] ‘into the [i] territory ... no matter whether the [g] precedes or follows the vowel. (Other spectrograms show that [k] and [ŋ] do the same.) ... Our data are very sketchy, but they seem to show that the beginning or end of a vowel sound, where it is in contact with a consonant, is modified in a manner characteristic of that consonant.‘
14 It is interesting to note that our accented [a] bears somewhat the same relation to unaccented [a] that the ideal [a] bears to the ideal [a].