Computer Sound Transformation

http://www.trevorwishart.co.uk/transformation.html#F7
Spectral cleaning was developed using a comparative method - part of the spectrum deemed to be (mainly) noise (and, in some options, part of the spectrum deemed to be clear signal) being compared with the rest of the signal and appropriate subtractions of data or other modifications made.

From a musical point of view, the most innovative early new developments were spectral banding, a rather complicated 'filter', which enabled the spectrum to be divided into bands, and various simple amplitude-varying (and in fact frequency-shifting) processes to be applied to the bands, spectral tracing and spectral blurring.

Spectral tracing simply retains the N channels with the loudest (highest amplitude) data on a window-by-window basis. If N is set to c. 1/8th the number of channels used in the PVOC analysis, this can sometimes function as an effective noise reduction procedure (the value of N which works best depends on the signal). When N is much smaller than this, and a complex signal is processed, a different result transpires. The small number of PVOC channels selected by the process will vary from window to window. Individual partials will drop out, or suddenly appear, in this elect set. As a result, the output sound will present complex weaving melodies produced by the preserved partials as they enter (or leave) the elect set. This procedure is used in Tongues of Fire (14).

Spectral blurring is an analogous process in the time dimension. The change in frequency information over time is averaged - in fact, the frequency and amplitude data in the channels is sampled at each Nth window, and the frequency and amplitude data for intervening channels generated by simple interpolation. This leads to a blurring or 'washing out' of the spectral clarity of the source.

Arpeggiation of the spectrum (a procedure inspired by vocal synthesis examples used by Steve McAdams at IRCAM to demonstrate aural streaming) was produced by 'drawing' a low frequency simple waveform onto the spectrum. This oscillator rises and falls between two limit values - values of frequency in the original spectrum - specified by the user. Where this waveform crosses the spectral windows, the channel (or surrounding group of channels, or all the channels above, or all those below) is amplified. Spectral plucking was introduced to add further amplitude emphasis (and an element of time-decay of the emphasized data) to the selected channels.

timestretch or compression and its range
segment density and its range
segment size
segment transposition and its range
segment amplitude and its range
segment splice-length
segment spatial position
segment spatial scatter and its range
segment timing randomisation
segment search-range in the source