Marie K. Huffman,
(12/2-12/5Town & Country Hotel, 619-291-7131)
Elyse Tamberino
Department of Linguistics
State University of New York
Stony Brook, NY 11794-4376
Popular version of paper 2aSC21
Presented Tuesday morning, December 2, 1997
134th ASA Meeting, San Diego, CA
Embargoed until December 2, 1997
Acoustic descriptions of vowels have many uses. For example, speech pathologists, educators and developers of speech-based technologies want to know what are typical examples of vowel sounds produced by healthy individuals. They also need to know how representative such examples are--in other words, how much variation is there between speakers? Some vowel variation arises from simple differences in anatomy between individuals. Other variation may be due to regular, consistent differences in pronunciation between groups of individuals--what we think of as dialect. Clearly, research and applications involving speech must take differences in dialect into account. Our research aims to extend existing knowledge of the acoustic differences between the vowels in American English dialects by providing an acoustic description of the vowels of one dialect of English spoken in central Long Island.
The nature or quality of a vowel is determined by the vocal tract shape--a combination of the position and shape of the tongue and lips, and the position of the soft palate. The differences in vocal tract shape between vowels mean that for different vowels, the air in the vocal tract vibrates in different ways. Each vowel's characteristic acoustic "signature" may be described as the combination of several resonances, or preferred frequencies, of air vibration. By measuring these frequencies, referred to as formants, we can produce a quantitative description of the way vowels sound--how they are similar, and how they are different. The most common method of studying vowel formants, and the methodology adopted here, is to measure the two lowest formants, and to compare where vowels fall in the "space" defined by these two dimensions. The formants also give a rough indication of the physical space in the mouth used for vowel production--the lowest formant gives an indication of how high in the mouth the tongue is during production of the vowel, and the second formant gives an indication of how far front in the mouth the tongue is during production of the vowel.
A crucial question which arises when studying vowels is the problem of where in time to measure the vowel formants--the beginning of the vowel? the middle? the end? A vowel is rarely a static, unchanging quality, like the beep of an answering machine. Instead, the tongue and/or lips are moving during vowel production, producing a dynamic sound quality. A number of previous studies have shown that in general, the acoustic nature of vowels, and the way listeners hear vowels, can be best described by trying to characterize this dynamism, by measuring the formants at more than one timepoint in the vowel. Our study examines the relative importance of time-varying formant change in characterizing vowel differences in Long Island English.
Many American English speakers have five distinct vowels produced in the front of the mouth (the vowel sounds in "beet," "bit," "bait," "bet" and "bat"). However, speakers on Long Island and in some other regions have an additional vowel, which is produced with a higher jaw and tongue position than the vowel in "bat." Thus, while many American English speakers have the same basic vowel quality in "bat" and "bad" (spkr1.wav), many Long Island English speakers have different vowels in these words, with the higher vowel occuring in "bad" (spkr2.wav). If Long Island speakers have a greater number of vowels produced in the front of the mouth (which is a finite space), this could mean that it is relatively harder to keep Long Island vowels consistently different in acoustic quality. If vowels are not kept distinct in quality, listener confusions may result. How, then, do speakers of Long Island English keep their front vowels distinct? What acoustic properties might help listeners tell these vowels apart? We hypothesized that in such a case, where there are several vowels produced in the same general part of the mouth, the dynamic quality of the vowels--the way the formants change over time, could be especially critical to distinguishing the vowel qualities.
We recorded mono-lingual English speakers from Long Island saying words that had the same consonant sounds, but contained different vowel sounds, such as "heed," "hid", "head" and "had". Using computerized speech analysis software, we identified and measured the first two formants of the vowels in these words. Following methodology of a recent study of upper midwestern English (J. Hillenbrand et al. [J. Acoust. Soc. Am. 57, 3099-3111], we took our formant measurements at three time points in each vowel: 20% 50% and 80% of the vowel duration. Using a statistical technique called discriminant analysis, we then evaluated how the vowel formant values of different words compared at these time points. Our results for Long Island English support the general finding of the earlier study, which is that if we describe each vowel with the formant values measured at only one time point_the 50% point or midpoint, the vowels occurring in different words, such as "bed" versus "bad," do not appear to be that different. However, if we characterize each vowel by reference to the formant values at two time points_the 20% and 80% points, we can more accurately determine, from formant values, which word was said. Vowel classification for our female voices improved overall by 11 percentage points when time-varying formant values were considered. Our own specific hypothesis was also confirmed, which is that including this time-varying description of vowel formants was particularly critical to accurately distinguishing among the front vowels of the Long Island speakers, vowels like those in "bed" and "bad". Here, analysis using formant values at only the vowel midpoint resulted in 40% of the vowels in "bad" being misclassified as being instances of the vowel in "bed," while incorporating formant values at the two timepoints resulted in near perfect classification of the vowels in these words.
In summary, having an additional vowel, the vowel in "bad" makes the set of front vowels more crowded for Long Island speakers, and one way to keep the vowels distinct is to use the space available in different ways: two vowels produced in the same part of the mouth may nonetheless be quite distinct in quality because one is produced with a slight rising or lowering movement of the tongue (changing the first formant over time), while another may be produced with a slight backing or fronting movement of the tongue (changing the second formant over time). A time-varying description of vowels provides a better account of the critical properties of vowels which make it possible for listeners to tell them apart.