posted 17 September 2000 06:06            

I have been fiddling with the various parameters of the FormantBankOscillator Sound to gain an understanding of what it really does.

So far I have found that the Sound is really just a retriggerable wavetable oscillator followed by a summed bank of "formant" filters. The FormantImpulse wavetable appears to be read out as wave samples at the sample rate.

The default settings for the Spectrum produce sometimes dissonant bell-like sounds, and it appears that the bandwidth of the formant filters is producing this dissonance by the extended ringing of inharmonic filters whenever their randomized bandwidths go near zero or negative. The bandwidth must be clipped to some small value near zero and not allowed to go negative.

The NbrImpulses parameter establishes a limit on the number of concurrently sounding wavetable oscillators. With a rapid impulse stream some pileup of these oscillators undoubtedly occurs and you have some algorithm for either cutting them short and starting up new ones from the beginning of the FormantImpulse wavetable, or else you ignore additional incoming pulses until an oscillator is freed up. It sounds to me like a combination of these two allocation policies is occuring.

Can you offer any enlightenment on the inner workings?

Thanks for such a creative instrument!

posted 17 September 2000 13:57            

When you specify a wide bandwidth (relative to the frequency of the FBO), the impulse responses are short and don't overlap; you can get by with roughly the same number of impulses as formants.

When you specify a narrow bandwidth, the impulse responses are long and overlap; in this case you may need to have NbrImpulses much larger than NbrFormants.

If all of the impulses are in use when a new one tries to start up, the new impulses are ignored.

posted 17 September 2000 23:41            

One of the confusing things about the docs on the FormantBankOscillator was the description about impulse responses. The document implies that each formant filter has the same impulse response. But the impulse response itself has spectral content. Hence, either there is one impulse response per impulse going through a bank of parallel formant filters, or else the same impulse response is used in each filter but read out at different sample rates according to the filter center frequencies. Here is where a block diagram or an equation would really come in handy!