Analysis of Tongue Parameters for Vowels
Lloyd Rice · The Journal of the Acoustical Society of America · 1972
A computer system is described that will accept a spoken vowel as input and then adjust the area function of a vocal-tract synthesizer to duplicate the input vowel quality as nearly as possible. The synthesizer output is matched against the input vowel spectrum using the Markel inverse filtering system. The program allows the definition of alternative parameter models for controlling the vocal-tract area function and represents a preliminary attempt to simulate the major vocal articulators. Each of three to six parameters is defined in terms of its contribution to the movement of each of the 18 sections used in the line analog synthesizer. Thus, the action of parameter Pi is defined by the 18-element vector of weights Wi1, Wi2, ⋯, Wi18. Any vocal tract shape within the range of the model is specified by assigning the appropriate value Pi to each of the parameters. The deflection of the jth section away from the neutral tract position is then given by the sum of products P1W1j+P2W2j+P3W3j+⋯. Various parameter models will be explored to determine their usefulness in providing unique and reasonable tongue shape descriptions.