[sdiy] Formant filters, yay!

mskala at ansuz.sooke.bc.ca mskala at ansuz.sooke.bc.ca
Thu Mar 9 19:54:30 CET 2017

On Thu, 9 Mar 2017, Richie Burnett wrote:
> Also, if those "Amp (dB)" values are the amplitudes of the formant peaks in
> the frequency spectrum, then they may or may not take into account what the
> excitation waveform was.  If you go about making a filter bank and drive it
> from a sawtooth oscillator for example, the sawtooth waveform will already
> have a -6dB/oct rolloff to it's spectrum before you start boosting and cutting
> areas with the formant filters.  Something to bear in mind.

I'm sure they describe the *signal*, not the *filter* - so they represent
the actual amplitudes of the frequency components, not the attenuation
required from some specific excitation waveform.  They're from the Csound
manual, and someone synthesizing formants in Csound would not necessarily
do it by filtering a single excitation waveform in the first place.  I
just checked the Csound manual for the original source, which I thought
was cited there, but it doesn't seem to be.  Nonetheless, I'm pretty sure
that the numbers came from a scientific study that actually measured the
frequencies in human voice spectrograms, and that could explain the narrow
bandwidths.  The numbers would've come from actual voices, not from
someone's idea of what is easy to build into an electronic filter.

This does mean that someone who wants to use those coefficients to build a
filter and run it with something other than white noise as input, would
need to adjust the higher-frequency coefficients accordingly (less
attenuation because the input signal already has reduced amplitude at
high frequencies).

Matthew Skala
mskala at ansuz.sooke.bc.ca                 People before principles.

More information about the Synth-diy mailing list