Introduction to Spectral Processing
FREQUENCY ANALYSIS FILES
The CDP Spectral functions operate in the Frequency Domain, or Spectral Domain, processing mainly Frequency Analysis files (.ana), derived from sound files via the Phase Vocoder (PVOC ANAL). Other processes manipulate pitch data extracted from Frequency Analysis files (see REPITCH).
SPECTRUM and SPECTRAL ENVELOPE
A sound may be analysed into a number of component frequencies (partials) of varying amplitude:
The image shows a single spectrum from a single time-slice, since sound usually evolves through time. The display shows a clear amplitude curve, from low to high frequencies, known as the spectral envelope. A frequency analysis file consists of a series of spectra from overlapping time-slices, or analysis windows, in which both the spectral envelope and the frequencies may (and normally do) vary over time. The frequency content, and the relative amplitude of the partials, are largely what gives a sound its distinctive timbre, or "tone-colour".
The Phase Vocoder's FFT analysis divides the frequency spectrum into a number of channels (frequency bands, also known as "bins"). Both the frequency information and the amplitude of each channel can be altered while in the spectral domain (see below). Or the sound can be partititioned into separate sets of channels (SLICE), to be treated independently prior to a possible remix.
OPERATIONS ON SPECTRAL AMPLITUDE
Many spectral processes transform the spectral envelope or other aspects of amplitude. Sounds can be successfully morphed from one to another (e.g. MORPH, NEWMORPH) or you can make hybrids by imposing the characteristics of one sound on another (VOCODE, CROSS, SPECSPHINX). Other processes manipulate how the spectrum changes across time, e.g. by blurring (BLUR), freezing (FREEZE, HOLD, MAGNIFY) or time-stretching the sound (STRETCH TIME). Some take a drunken walk along the analysis windows (DRUNK), weave a path through them (WEAVE) or randomize their order (RAND). The amplitude in each channel can be prolonged over time to blur its evolution (SUPERACCU).
An important group of spectral processes deals with channel amplitude and spectral peaks, exaggerating the peaks to focus the sound (EXAG, FOCUS), or raising the level of weaker bands to saturate the spectrum with noise (NOISE, SPREAD). Filtering is especially successful in the spectral domain (FILTER), as unwanted frequencies can simply be eliminated, for example, to clean up the sound (SPECNU CLEAN). We can retain (TRACE) or remove (SUPPRESS) the strongest frequencies, or impose the formant characteristics of vowels (VOWELS).
OPERATIONS ON FREQUENCY
The frequency information can be transformed directly. Pitch can be transposed without affecting time (TRANSPOSE(F), TRANSP, OCTMOVE), or the spectral frequencies can be stretched (STRETCH SPECTRUM) or shifted up or down (SHIFT), creating inharmonic spectra out of harmonic ones, or even turned upside down (SLICE, Mode 5).
OPERATIONS ON PITCH
Pitch is perceived when the partials are in a harmonic relationship (integer multiples of a fundamental frequency). The REPITCH programs can extract pitch, if it is present, either as a binary pitch data file (.frq), or as a breakpoint text file which can be easily altered.
Several functions are available to transform pitch data or combine pitch files to make a transposition file (.trn). The transition back to sound is via COMBINE MAKE, which combines a pitch data file (binary or breakpoint) with a spectral envelope, or formant file (.for). However, any alterations to pitch are only audible to the extent that the frequencies are also present in the formant file.