The waveforms named "NoiseBandXXX" define a non-harmonic sound band (even though it usually has a perceived pitch) centered on the sound's frequency.
Two sliders control the parameters of this waveform:
spread: the range (from 0 to +/-80Hz) over which the noise band is formed.
density: the density at which the noise band is filled. 0 means only one sine is generated, 1 means 240 parallel sine waves are generated across the spectrum.
The "CrazyNoise" examples are variations of this noise band. They were originally programming mistakes, but the sound they produced was found sufficiently interesting to keep them in.
The waveforms named ChamberlinNoiseBand and ButterworthNoiseBand are subtractive synthesis-based noise bands. Their effect is quite similar to the additive noise bands, but they are much more efficient to compute, specially for larger bands (above 20 Hz).
the Chamberlin Noise Band creates a noise band by applying a Chamberlin 12dB/Oct pass-band filter to a white noise signal. It has two parameters: resonance sets the resonance of the filter, thus approximating the width of the pass band. The distortion parameter introduces a bit of of white noise to reduce the slope of the filter. The technical implementation of the filter can be found at: Hal Chamberlin, "Musical Applications of Microprocessors," 2nd Ed, Hayden Book Company 1985. pp 490-492.
the Butterworth Noise Band creates a noise band by applying a Butterworth 12dB/Oct pass-band filter to a white noise signal. It has two parameters: resonance sets the resonance of the filter, thus approximating the width of the pass band. The width parameter sets the width of the pass band. This filter is a band-pass version of this algorithm. There is a big likelyhood that this algorithm will need to be improved, meaning future versions of the software may not sound exactly alike.