Audio-rate filter modulation
Most plugins have two distinct processing sample rates
Imagine a sine LFO smoothly sweeping a filter’s cutoff up and down. When the LFO is cycling around at a speed below 20Hz, we perceive this low-frequency oscillation as an obvious wobble of the filter’s frequency. Speed the LFO rate up into audio-rate territories, however, and the wobble increasingly blurs, introducing complex harmonic overtones and noise-like dissonance into the filtered signal. This is akin to FM synthesis, whereby the frequency (pitch) of one oscillator is used to modulate another’s for a similar sort of harmonic complexity. As it’s such a dissonant effect, audio-rate filter modulation is usually best used for trashy drum-mangling, sci-fi FX design and lo-fi degradation.
Cytomic head honcho Andrew Simper goes deeper into the complexities of coding this type of rapid filter movement. “Most plugins have two distinct processing sample rates: one for the audio, and a slower one for the modulation. This slower rate is usually called ‘control rate’, and it is typical that this is around 64 times slower than the sample rate. This means that there is only one new control value – for example, one new cutoff frequency for the filter – for every 64 samples of audio processed. If the audio sample rate is 44.1 kHz, and modulation is processed at 64 samples, then the maximum frequency that can be represented is around 700Hz. This is a reasonable rate for slow modulation, but for fast modulation the resultant sound is very harsh.
“Audio-rate modulation isn’t much use unless all the modulation sources produce their anti-aliased output at audio rate as well. Since there is no difference between an audio signal and a control signal, this is as close to the analogue world as possible. Unfortunately, this speed of update can be dangerous for many common digital filter algorithms – it can crash and sound terrible.
“In The Drop, all modulation sources and destinations are calculated at audio rate, the modulators are anti-aliased, and the filters all preserve the structure of the circuit and position and shape of all major nonlinearities, so everything sounds ultra smooth and as ‘analogue’ as possible. This does take more CPU, but you can push it much harder.”