| |
Abstract:
In this paper, the acoustic features of speech characterized
by 3-D vocal tract shapes are investigated with a simulation
method of a 3-D finite element method(FEM). We examine the
effects of the asymmetrical shape of the vocal tract using a
simulation model which has a cascaded structure of 36 cross
sections based on MRI data of the vocal tract for the Japanese
vowel /a/ of an adult male. Each cross section of the model is
obtained by converting the MRI data into foursquare elements of
equal size. The cross-sectional shape of the simulation model
corresponds well with that of the original MRI data. Each element
of the model is in the shape of a cube and the same size. The
utilization of the cube elements facilitates to construct the
model. Furthermore a bent version of the above straight
configuration is constructed in agreement with the vocal tract
shape. In the case of the bent configuration, all the elements
are not the same cubical shape. In these models, the 3-D
radiation model with a radius of 4cm is attached to the radiation
end. A rigid wall is assumed. The driving surface is driven by
the sine waves. Vocal tract transfer functions are computed from
the simulation results. For a purpose of comparison, the vocal
tract transfer functions for our FEM models with an elliptical
cross-section are shown. The results show that the first three
formant frequencies of the vocal tract transfer functions move to
lower frequencies compared to those of the simulation model with
the elliptical shape. Especially in the case of the bent
configuration, the shift of the third formant frequency becomes
larger than that of the straight configuration. This fact
suggests that the asymmetrical shape of the vocal tract has an
effect on the length of the acoustic vocal tract, which becomes
longer in length than the real length. In the VTTFs some sharp
peaks appeared at higher frequencies above 3kHz, which might be
effects of the asymmetrical shape or the effects of the
discontinuous shape of the boundary elements.
|
