Estimation of the vocal-tract length from acoustic data
H. Wakita · The Journal of the Acoustical Society of America · 1974
There are several ways to estimate the vocal-tract length from acoustic data. The use of the higher formant frequencies is one of them, but this does not give satisfactory results. The use of the zeros and poles of the lip impedance [Paige and Zue, IEEE AU-18, 265–270 (1970)] is not quite accurate because of the approximation involved. A third choice which is the topic of this paper is to estimate the length using formant frequencies and bandwidths. In this case, the tract length is chosen such that the error criterion proposed by Paige and Zue is minimized. This was not an easy task until a simple transformation from given acoustic data into a unique tract shape was developed [Wakita, SCRL Monograph No. 9 (1972)]. Utilization of the first four formant frequencies and bandwidths resulted in errors of less than 5% in estimating the length when tested on the tract shapes obtained by Fant. The length information can be used for obtaining a more reasonable vocal-tract shape from speech sounds and for normalizing speech parameters among various speakers. [The Office of Naval Research supported this work.]