On Sat, Jul 13, 2013 at 05:03:59PM +0100, Harry van Haaren wrote:
A) Create different wavetables for each octave. Base
octave includes all
harmonics. Octave 1 has the top half of the harmonics removed by FFT. Oct 2
has the harmonic content halved again.
Fine for fixed frequencies (aeolus works this way but not per octave,
all notes are precomputed separately), but becomes problematic if the
you allow frequency modulation over a wide range.
Can be combined with oversampling to make interpolation easier.
B) Oversample the single waveform x8. Play the
oversampled audio back, and
lowpass with a steep rolloff just below the nyquist of the output
samplerate. Removal of the otherwise aliasing harmonics is done at the
higher samplerate, so its not aliased yet at that stage.
This could be done very efficiently, in particular if you accept some
compromises for the lowpass filter, and even more if the waveforms can
be summed before downsampling.
There are other methods, depending on what sort of waveforms
you want to generate.
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)