Madde works using additive synthesis, so that the output
signal is made by summing up sinusoids of harmonic ratio frequencies,
each one having its own amplitude. In a similar system, there are
no filters (strictly speaking) because spectral shapes (formants
included) can be easily obtained modifying in a suitable way the
amplitude of each sinusoid.
A glance to the Madde main Window.
In the upper left corner you can find the
If you "switch off" every formant (unchecking the
corresponding checkbutton) you can ear (and see in RTSect) the
simulated glottal signal: a sawtooth).
Beneath every formant you can read / modify its
frequency. You can move frequencies, all together by a common factor,
writing up this figure into the "Factor" entry. If you write here 1.3,
you will raise every frequency by 30%, If you write 0.8, you will pull
down every frequency by 20%.
Attention, please: you must make use of
the comma (",") or the dot (".") as a decimal separator, as a function
of what you have declared as your locale in Windows. Try to fix it.
You can find below the entries for the Qs. These are
"merit factors" of the "formant filters" (so speaking, in additive
synthesis you don't need actual filters). Higher values mean more
selectivity, lower Q values mean less selectivity. If you pull down
these values, you can ear the "formant effect" becoming less and less
prominent, till it disappear.
You can find the bandwidth beneath the entries, and you
can observe how they change when moving Q. A bandwidth of, to say, 80.0
Hz at 800 Hz of peak value means that @ 800-80 HZ and @ 800+80 Hz the
gain of the filter is 3 dB beneath the peak value.
This Q (or bandwidth) have a physical meaning, which
will become clear when studying the dissipative 2nd order systems (or
dissipative oscillators, either mechanic or electric). This factor is
tied to the dissipation in the system (friction, conversion from
mechanical/electrical energy to heat).
The F0 pane, left lower corner:
F0 (or f0) stay for the pitch (or fundamental
frequency, in this case), namely is a note (whose name you can find ad
the right of the frequency, in this case C4).
Madde makes use of many of the known features of the
simulated singing voice (see the works of Chowning, Sundberg and Rodet
on this subject). So the glottal signal is submitted to a random
frequency modulation (Flutter) plus a periodical (sinusoidal) frequency
modulation (Vibrato). You can set the parameters for these modulations:
Amplitude and frequency. This last one has a different meaning for
flutter and vibrato. The flutter is a noise filtered around a frequency
(you can thus set this frequency, and the Q of the filter). The Vibrato
is a sinusoid, so that the frequency is shortly its frequency. The
amplitude (for both) is actually a Modulation Index (FM).
Then the partials (harmonic) pane:
Here you can find the slope of the amplitudes as a
function of the frequency. -6 dB/octave is exactly the slope which will
produce a sawtooth. If you modify the slope, you can reset the
original value by pressing Flat!.
In the windows below, you can further modify (or shape)
the formants, relatively to the chosen slope. Every red dot can be
moved individually, but you can also draw the shape by moving the
You can also modify the power of the signal (Level in
dB), and the number of partials (by default 40, which is quite enough
for the purpose)
Now, a very interesting window (menu "Settings", submenu
"Show F1/F2 map"):
If you draw using the mouse on the window, you can
modify both F1 and F2, thus modifying the pronounced vowel.