Determination of the Geometry of the Human Vocal Tract by Acoustic Measurements

Abstract

The geometry of the human vocal tract from the glottis (vocal cords) to the lips is one of the determining factors in human speech production and basic to a better understanding of articulation. Measurement of the vocal-tract geometry is difficult, especially that of its time-dependent behavior during connected speech utterances. Practically the only tools that have been available in the past for such measurements are radiography and cine radiography. Unfortunately, radiographic measurements and their evaluation are laborious. In addition, adequate cine radiography for longer utterances poses severe problems owing to dosage limitations. Alternative methods of obtaining information on the vocal-tract geometry have therefore always remained a desideratum. Some of the desired vocal-tract-shape information can be obtained by purely acoustic measurements. Two kinds of acoustic methods are discussed in some detail measurement of the eigenfrequencies (formants) of the vocal tract and measurement of the acoustic impedance at the lips. Both measurements allow the determination of an approximation to the cross-sectional area of the vocal tract as a function of distance along its axis. More specifically, if the logarithm of the area function is described by a Fourier series, these measurements allow the determination of certain terms of the Fourier series. The number of terms, and therfore the spatial resolution of the approximation, increases with the number of measured eigenfrequencies of the tract or with the number of measured singularities of the impedance function. Both methods are applicable to dynamic measurements during nonstationary utterances and hold promise for supplying large amounts of area data necessary for specifying or refining articulatory models of speech production. The results can be portrayed by computer-drawn single frames or motion pictures of the acoustically derived area functions.