Third International Conference on Spoken Language Processing (ICSLP 94)
A neural network was used to estimate the lateral shape of a tongue from speech. An experimental result indicates the PARCOR coefficients as being the best parameter. The average error attained in the estimation is 1.2 mm in rms. For speaker adaptation, ultrasonic image of the dorsal part of the tongue was used and the result verifies its effectiveness.
Bibliographic reference. Hiraike, Masahiro / Shimizu, Shigehisa / Mizutani, Takao / Hashimoto, Kiyoshi (1994): "Estimation of the lateral shape of a tongue from speech", In ICSLP-1994, 591-594.