Some Measurements on the Vowel Sounds of Conversational Speech
John Swaffield, J. N. Shearme, John N. Holmes · The Journal of the Acoustical Society of America · 1961
Fairly well-defined and generally accepted regions of the (first formant frequency)—(second formant frequency) plane (referred to as the F1F2 plane) can be associated with each speech vowel sound. Delineation of the boundaries of these regions may be based on analysis of spoken vowels or on perception of synthesized vowels; in either case, the data are derived from the study of isolated vowels or monosyllables, and the resulting two sets of regions prove to be roughly the same. An alternative study, described in the paper, can be made, however, based instead on the vowel sounds occurring in “connected” or “conversational” speech. In this case the density of distribution of points within the F1F2 plane shows that there is little or no clustering of stationary points within the original (“monosyllable”) vowel regions and certainly permits neither identification of old, nor definition of new, regions. A set of tracks can be constructed, however, to correspond to the almost continuously moving F1F2 points and from these can be derived, by suitable processing, a new set of regions of simple shape, one region for each vowel sound as before. The differences between the set of regions for “monosyllable” vowels, i.e., the generally accepted set, and the set for “conversational” vowels, which may be reasonably regarded as the set in normal use, are striking. Two differences are (a) the set of vowel regions is now clustered very much more closely together than before and (b) for a given speaker the positions of “monosyllable” vowels do not even lie within the regions of the corresponding “conversational” vowels.