How to do formant
synthesis?
We can use source/filter models to synthesize
the human vocal systems. In voiced phonation, glottal folds open and close
periodically, producing pulse-like excitation. This glottal pulse source
then is filtered by the vocal tract, which forms a resonating system. The
following is a simple matlab code to simulate source/filter model.
formant_synth.m
Figure 1 shows the waveform and the spectrum of band-limited
impulse train as our source model. Figure 2 shows the spectrum of the filter
with three resonant peaks that correspond to the first three formant frequencies
of /ah/ sound, and Figure 3 shows the waveform and the spectrum of resulting
output signal.
<Figure 1> Waveform and spectrum of band-limited
impulse train as a source model.
<Figure 2> Frequency response of the
filter model with three resonant peak frequencies.
<Figure 3> Output signal and its spectrum.