You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
49 lines
2.7 KiB
49 lines
2.7 KiB
The Klatt formant synthesizer
|
|
|
|
sociated with the cavity in front of the oral constriction are strongly excited. The
|
|
theory is supported by the formant amplitude specifications for fricatives and
|
|
plosive bursts presented in Chapter 11. These amplitude control data were derived
|
|
from attempts to match natural frication spectra.
|
|
|
|
There are six formant resonators in the parallel configuration of Figure 12-6.
|
|
A sixth formant has been added to the parallel branch specifically for the synthesis
|
|
of very-high-frequency noise in ss, zz. The main energy concentration in these
|
|
alveolar fricatives is centered on a frequency of about 6 kHz. This is above the
|
|
highest frequency (5 kHz) that can be synthesized in a 10,000 sample/second
|
|
simulation. However, in a ss, there is gradually increasing frication noise in the
|
|
frequencies immediately below 5 kHz due to the low-frequency skirt of the 6 kHz
|
|
formant resonance, and this noise spectrum can be approximated quite well by a
|
|
resonator positioned at about 4900 Hz. We have found it better to include an extra
|
|
resonator to simulate high-frequency noise than to move F5 up in frequency when-
|
|
ever a sibilant is to be synthesized, because clicks and moving energy concentra-
|
|
tions are thereby avoided.
|
|
|
|
Also included in the parallel vocal tract model is a bypass path. The bypass
|
|
path with amplitude control AB is present because the transfer function contains
|
|
no prominent resonant peaks during the production of FFr, vv, TH, and DH, and
|
|
the synthesizer should include a means of bypassing all of the resonators to
|
|
produce a flat transfer function.
|
|
|
|
During the production of a voiced fricative, there are two active sources of
|
|
sound, one located at the glottis (voicing) and one at a constriction in the vocal
|
|
tract (frication). The output of the quasi-sinusoidal voicing source is sent through
|
|
the cascade vocal tract model, while the frication source excites the parallel branch
|
|
to generate a voiced fricative.
|
|
|
|
12.2.7 Simulation of the cascade configuration by the parallel configuration
|
|
|
|
The transfer function of the laryngeally excited vocal tract can also be ap-
|
|
proximated by five digital formant resonators connected in parallel. The same
|
|
resonators that form the parallel branch for frication excitation can be used if
|
|
suitable values are chosen for the formant amplitude controls.
|
|
|
|
The following rules summarize what happens to formant amplitudes in the
|
|
transfer function T(f) of a cascade model as the lowest five formant frequencies
|
|
and bandwidths are changed. These relationships follow directly from Equation 6,
|
|
under the assumption that each formant frequency F(n) is at least five to ten times
|
|
as large as the formant bandwidth BW(n):
|
|
|
|
1. The formant peaks in the transfer function are equal for the case
|
|
|
|
145
|