Locating Acoustic Nuclei

The methods used here to choose a time slice of the acoustic vowel as representative of the vowel's nucleus are similar to those of Labov, Yaeger, and Steiner (1972:29) (LYS). Typically the maximum of the F1 contour within the acoustic vowel is chosen. This time-point reflects the point of maximum opening of the mouth (cf. Chapter 2, page ), and also the target or point of inflection of the opening gesture of the vocal tract which articulatorily implements a phonological vowel. As the vocal tract moves from consonant to (simple) vowel to consonant, it opens and closes. If there is an articulatory or acoustic target associated with a vowel, the point of maximum opening of the mouth is commonly the point at which that target is most closely realized. There may be no actual target or ``magic moment'' within a syllable that constitutes the essential realization of the vowel nucleus; instead the entire trajectory of the gesture associated with a sound may be an essential part of the sound. Even here, the point of maximum opening is of great importance, since these trajectories may be most simply characterized by describing the acoustic or articulatory configuration at this point, along with the time-course of the transition to this point from preceding segments and from this point to following segments. This is the technique of ``temporal decomposition'' (Atal 1983), a mathematical method in which the actual smooth sequence of overlapping gestures is decomposed into a sequence of separate targets and a set of time functions which describe how they overlap and shade off gradually from one into the next. This method may be useful to gesture theorists (cf. Browman & Goldstein 1986) in the mathematical characterization of the entire trajectories of speech gestures.

It is easy to delimit and listen to very short pieces of the speech waveform. I often listened to a short segments (as little as 30ms) which included the time-slice marked as the acoustic nucleus of the vowel in question, as well as larger segments containing the entire vowel and adjacent consonants. It was my frequent impression that the short nuclear segment sounded most like the overall impressionistic quality of the entire vowel, as compared with earlier or later short segments. Even segments adjacent to the acoustic nucleus in lengthened vowels did not generally sound as much like the entire vowel as the segment centered on the F1 maximum. Thus the F1 maximum seems to be an auditorily important point in the course of the acoustic vowel.

If F1 has no maximum, then the point of inflection of F2 (maximum or minimum) is chosen, considering that the gesture that created the sound reached an inflection point at that time-point, and began to move in the direction of other targets. If neither has a point of inflection, then the mid-point of the acoustic vowel is chosen. These considerations must be altered in the case of downward-gliding vowels, such as ingliding, raised /æ/ [e], /ir, i:l/ [i, i], Jamaican /uo/, and the like which show formant trajectories in which F1 rises and F2 falls between the part of the acoustic vowel that corresponds to the vowel nucleus and the part that corresponds to the following glide. In such cases, the middle of the section that corresponds to the phonological nucleus of the vowel is chosen as the acoustic nucleus; commonly this is a steady-state preceding the offglide.

Labov, Yaeger, and Steiner (1972) considered only stressed monosyllables, while this study includes all vowels occurring in a lengthy discourse, from the shortest to the longest, in order to include all vowels, stress effects, and consonant contexts. Thus the restriction in LYS that the measurement be at least 50ms after the beginning of the syllable was abandoned here. If the acoustic vowel was 50ms or less in duration, the midpoint of the vowel was chosen.