Figure 51.2
Syrinx model.
574
Birds
DSP Implementation
Here is the core of our birdcall synthesiser (fig. 51.2). Starting on the left, there
are two inlets from the bronchi that will carry pulses into the ring modulator.
A third inlet is the ring modulator balance that varies between the sum and
difference of the incoming waves. Output from the ring modulator is scaled by
the signal value on the fourth inlet and added to an offset (fifth inlet), giving us
the base frequency of the bottom pulse oscillator. We can vary the pulse width
of this from the last inlet.
Figure 51.3
Pulse wave oscillator.
Three pulse oscillators, seen as
vposc
in the
parent patches and shown in figure 51.3 opposite,
are used. One is used for each bronchus/labia port
and one for the syrinx simulation. Frequency is set
by the first inlet and pulse width by the second. We
tend to use small deviations from a raised cosine
here, with values on the pulse width inlet ranging
between 1
.
0 (raised cosine) and 3
.
0 (slightly nar-
rowed pulse). The result of course is not centred
on zero, so both inputs to the ring modulator have
a DC offset that propagates through the patch. To
simplify matters we have not recentred the pulses;
the DC component is dealt with by filters later in
the chain.
Figure 51.4
Ring modulator.
For convenience the ring modulator is given here
again (fig. 51.4). Recall from the chapter on shaping
that multiplying two signals moves the harmonics
by their difference and adding them performs simple
superposition that leaves them unaltered but com-
bined. For two harmonics at the inlets we can expect
a mixture of potentially four harmonics at the out-
put, depending on the value of the crossfader. A
value of 0
.
0 at the right-hand crossfader inlet pro-
duces only difference components, while a value of
1
.
0 gives only the sum. At 0
.
5 we get an equal mix-
ture of both.
Finally, here are two filters in parallel for the trachea (fig. 51.5). These focus
the driving oscillation of the AM/FM syrinx into a narrower band. We can
expect some undesirable modulation sidebands, especially at low frequencies
where they may be reflected around zero. Fortunately the deviations needed
are mild so there won’t be any problems with sidebands getting above the
Nyquist and folding over at the top. A waveguide model for the trachea would
be nice, but most of the character we are interested in comes from the syrinx,
and so any realism advantage gained by a more sophisticated output tract might
be too subtle to warrant the extra cost. Resonance of this component remains
fixed during each call and should be between 2
.
0 and 5
.
0.
Do'stlaringiz bilan baham: |