Principal components of vocal-tract area functions and inversion of vowels by linear regression of cepstrum coefficients

A method for estimating vocal-tract shape from a target speech spectrum

“…It is clear that the increase in estimation error, i.e., 0.03 cm 2 for area and 0.02 cm for vocal-tract length, was very small when using the speech spectrum as the target in the present method. It is also valuable to compare the estimation results of our method with those of the linear mapping method [20]. The estimation results reported in the literature [20] show that the area error is almost the same (0.367 cm 2 ) and the length error is slightly better (0.150 cm) than those for our method.…”

“…It is also valuable to compare the estimation results of our method with those of the linear mapping method [20]. The estimation results reported in the literature [20] show that the area error is almost the same (0.367 cm 2 ) and the length error is slightly better (0.150 cm) than those for our method. However, it should be noted that the dataset used in the linear mapping method and the number of principal components differed from those in our study.…”

“…Therefore, previous studies did not consider the underlying physical relationship between the vocal-tract shape and speech spectrum. Instead, the vocal-tract parameter was related to the acoustic parameter using an articulatory-acoustic database [18,19] or a linear [20] or nonlinear [21] mapping method. In the database search approach, a fine discretization of every vocal-tract parameter is needed to relate the articulatory and acoustic parameters accurately.…”

A comparison of vocal tract perturbation patterns based on statistical and acoustic considerations

“…Statistical analyses of collections of tongue configurations or complete vocal tract shapes (i.e., area functions) have revealed that a small number of canonical deformation patterns (variously referred to as factors, components, basis functions, or modes) can explain most of the variation in vocal tract shape during vowel production (Harshman et al, 1977;Shirai and Honda, 1977;Jackson, 1988;Johnson et al, 1993;Nix et al, 1996;Story and Titze, 1998;Hoole, 1999;Zheng et al, 2003;Iskarous, 2005;Story, 2005;Mokhtari et al, 2007). These deformation patterns tend to exhibit similarities in shape across speakers and are related to specific formant frequency patterns when superimposed on a mean or neutral vocal tract shape.…”

“…studies have also reported principal component analyses of area functions that resulted in modes shapes similar to those in Figs. 1(a) and 1(b) (Meyer et al, 1989;Yehia et al, 1996;Mokhtari et al, 2007).…”