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.

52 lines
2.3 KiB

From text to speech: The MITalk system
12.1.5 Parameter update rate
Control parameter values are updated every 5 msec. This is frequent enough to
mimic even the most rapid formant transitions and brief plosive bursts. If desired,
the program can be modified to update parameter values only every 10 msec with
relatively little decrease in output quality.
12.1.6 Digital resonators
The basic building block of the synthesizer is a digital resonator having the
properties illustrated in Figure 12-5. Two parameters are used to specify the input-
output characteristics of a resonator, the resonant (formant) frequency F and the
resonance bandwidth BW. In Figure 12-5, these values are 1000 Hz and 50 Hz,
respectively. Samples of the output of a digital resonator, y(nT), are computed
from the input sequence, x(nT), by the equation:
y(nT)=Ax(nT)+By (nT-T)+Cy (nT-2T) (1)
where y(nT-T) and y(nT-2T) are the previous two sample values of the output
sequence y(nT). The constants A, B, and C are related to the resonant frequency F
and the bandwidth BW of a resonator by the impulse-invariant transformation
(Gold and Rabiner, 1968):
C=—e2"BWT
B=2¢"™8WTcos(2nfT) )
A=1-B-C
The constant T is the reciprocal of the sampling rate and equals 0.0001 seconds in
the present 5-kHz simulation.
The values of the resonator control parameters F and BW are updated every 5
msec, causing the difference equation constants to change discretely in small steps
every 5 msec as an utterance is synthesized. Large, sudden changes to these con-
stants may introduce clicks and burps in the synthesizer output. Fortunately,
acoustic theory indicates that formant frequencies must always change slowly and
continuously, relative to the 5-msec update interval for control parameters.
A digital resonator is a second-order difference equation. The transfer func-
tion of a digital resonator has a sampled frequency response given by:
A
T(f)=————— 3)
1-Bz-1-Cz2
where z=62™7 j is the square root of —1, and f is frequency in Hz which ranges
from 0 to 5000 Hz. The transfer function has a (sampled) impulse response iden-
tical to a corresponding analog resonator circuit at sample times nT (Gold and
Rabiner, 1968). But the frequency responses of an analog and digital resonator are
not exactly the same, as seen in Figure 12-5.
128