The inverse problem for the vocal tract: numerical methods, acoustical experiments, and speech synthesis.

In this article we will describe our recent experiments on the estimation of vocal tract area functions from transient acoustical measurements at the lips. Since the theoretical basis for such measurements has been discussed in several earlier publications we will concentrate on the measurement technique and the numerical procedures used to estimate the area functions from the measurements. The two main accomplishments we will report on are (1) We are now able to make measurements and compute and display the area function about 18 times per second. Such a display, which has not been possible before, allow one to visualize the shape of one's vocal tract as one changes the position of one's articulators. This display could form the basis of a visual aid for the improvement of the speech of profoundly deaf persons. (2) We have synthesized intelligible though not yet high quality speech from several sentence-length sequences of measured area functions. To the best of our knowledge, this is the first instance of continuous speech synthesized from direct measurements of area functions.