Return to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
N – no. of adjacent channels – must be <= half the no. of channels in the infile.
N may vary over time
The averaging of energy information causes high energy data to cross channels. The result is the intrusion of timbral data into channels which previously didn't have that data. Test results led to a 'roughening' of the sound, though without going as far as distorting it.
Note that this process relates to channels, not to windows (as does the BLUR BLUR process). This means that the averaging directly affects the frequency components of the sound.
The energy (amplitude) in adjacent channels is averaged out over Average-span adjacent channels, without affecting the frequencies of those channels. The result is to broaden, or defocus, any energy peaks in the spectrum.
This reduces the frequency definition of the sound, which is why it is in the BLUR category and is, in effect, a type of filtering.
End of BLUR AVRGBLUR AVRG appears to be useful to enrich the timbre of a sound and make its texture rougher.
infile – input analysis file made with PVOC
outfile – output analysis file
blurring – the number of windows over which to average the spectrum
blurring may vary over time: provide the name of a time blurring breakpoint file for this parameter.
BLUR BLUR time-averages the spectrum. It 'blurs' detail in the time dimension by interpolating between the spectral envelope values of the start and end windows blurring windows. Note that it is not interpolating continuously over all the windows inbetween, just between the data in the start and end windows. The overall result is somewhat affected by just how different the data is in these two windows. The interpolation process produces a 'straight line' (linear) scale of values between the start and end points.
BLUR BLUR differs from HILITE BLTR in the absence of the 'trace' parameter. This means that the blurring occurs without any reduction in the number of partials.
As with all of these programs, the result is greatly affected by the nature of the input sound. The more windows blurred, the more blurring or smoothing of the sound you might expect to happen. However, you might not notice much difference if the sound is already constant (or similar at the start and end points). You will probably need a sound with a great deal of internal change for the blurring to have a perceptible effect.
The musical results of this process begin with a softening of the attack transients, so is highly effective with, for example, plucked or percussive sounds. A time-varying transition from the original to a very blurred effect can be achieved by a simple breakpoint file which sets a low value for blurring at the beginning of the sound and a high value at the end. It is possible for the sound to disappear gradually into its own ambience. This powerful technique can therefore useful in the creation of 'ambient' music, as well as any degree of softening of the original sound.
Also see: HILITE BLTR and HILITE TRACE.
End of BLUR BLUR
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
inanalfile – input analysis file
outanalfile – output analysis file
blurfact – time over which the upper part of the spectrum is blurred (Range: 0 to end-of-file)
blurabov – Frequency above which the spectrum is blurred (Range: 0 to sample-rate/2, e.g. 22050)
[-llocut] – Frequency below which bass is cut off. (Range: 0 to sample-rate/2; default: no bass cutoff)
CALTRAIN appears to be a variant of BLUR, affecting partials above a given frequency. The result is normally a roughening or distortion of the sound in its upper frequencies.
End of CALTRAIN
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
1 Randomise partial amplitudes
2 Randomise partial frequencies
3 Randomise partial frequencies upwards only
4 Randomise partial frequencies downwards only
5 Randomise partial amplitudes AND frequencies
6 Randomise partial amplitudes, and frequencies upwards only
7 Randomise partial amplitudes, and frequencies downwards only
infile – input analysis file made with PVOC
outfile – output analysis file
aspread – maximum random scatter of partial-amps (Range 1-1028)
fspread – maximum random scatter of partial-frqs (Range 1-4)
aspread and fspread may vary over time
Formerly SPECHORU, this process attempts to achieve a chorusing effect by randomising the amplitude and frequency values of the partials. It is based on an idea of Stephen McAdams. See his Spectral Fusion and the Creation of Auditory Images (1981).
If very large amplitude aspread values are used, the sound will turn to noise. The ranges shown for aspread and fspread appear to be rather extreme, but they are rescaled within the program. There is in fact no mathematical limit to the values which can be entered. In practice, the 'chorusing' effect itself is achieved by values just a little above 1!. Values of 2 or 3 begin to create a granular effect, and values of 10, 100 and 1000, for example, create more and more noise.
'Chorusing' is one of the standard effects used to enliven and enrich a sound. It becomes available here in an open-ended context, where the effect can be driven beyond the usual bounds into granulation and noise effects, useful when a certain textural or gritty quality is needed.
In Mode 5, a burbly but comprehensible vocal transformation may occur with aspread = 30 and fspread = 4.
End of BLUR CHORUS
Return to List of BLUR functions at top
of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
range –the maximum step (in windows) for the drunken walk: <= 64
starttime– the time (in seconds) in the file at which the walk should begin
duration– the required duration of the outfile after re-synthesis it may be longer than infile
-z eliminates zero steps (window-repeats) in the drunken walk
Beginning at starttime, a user-defined point in the infile, BLUR DRUNK moves through the file reading windows, but jumping either forwards or backwards, from one window to another, in a random way. The size of these random jumps cannot be greater than range. This random process is known as a drunken walk. It proceeds until the specified duration of the output sound has been generated.
If the process tries to leap right out of the soundfile at its end or its start, it is automatically reflected back into the sound near to the leap point. The process never reads backwards through the file, though it may leap backwards before starting its next read.
This process can be used to lurch about in a soundfile, producing a jumbled version of infile to varying degrees. The salient parameter is range. If range is small, the output will tend to linger around your start-point in the file, progressing very slowly away from it in an arbitrary direction. This results in a mix of time-stretching and slow wandering through the source. If range is large, the drunken walk tends to leap about wildly in the file, scrambling the source sound.
End of BLUR DRUNK
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
noise – Range 0 (no noise in spectrum) to 1 (spectrum saturated with noise)
noise may vary over time
This functions enables one to move a sound source towards pure noise, by making the data in every channel – most of which is actually low level noise – equally loud. Total saturation will reduce all sounds to a very similar noise signal. Partial or gradual saturation is possible, through the use of values less than 1 or by using a time-varying breakpoint file. These work in the usual way, with gradual change between different time points to which are assigned different noise values.
This technique can be used to cause sound material to emerge from obscurity to clarity, or v.vs.. Also, a carefully chosen noise factor can be used to colour a sound or soften its edges.
End of BLUR NOISE
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
keep – number of (randomly chosen) blocks to keep in each spectral window (Range: 1 to no. of chans.)
-bblocksize – frequency range of each block (default is width of 1 analysis channel. (Rounded internally to a multiple of channel width.)
-r number of blocks actually selected is randomised between 1 and keep
-n turn OFF normalisation of resulting sound
keep and blocksize may vary over time
This function throws away a specified proportion of the analysis data in each window. Unlike HILITE TRACE, the channels chosen to be suppressed are selected entirely at random. The spectrum is thus thinned out in an unpredictable fashion.
Remember that the number of channels in an analysis file is determined by the value for the -N flag of the Phase Vocoder (now referred to as -cpoints). For example, if it was -c1024, there will be 512 + 1 channels, each containing frequency and amplitude data.
Before scattering, the channels are gathered together into blocks of adjacent channels, and these blocks-of-channels are then either retained or discarded (at random).
Blocksize sets how many of these channels form one block. Keep sets how many of these blocks (i.e., groups of channels) will be randomly chosen and retained from the total of 513.
The values for keep and blocksize therefore need to be worked out in relation to the value given for -c. For example, retaining keep = 10 and blocksize = 6 will retain a total of 60 (randomly selected) channels from the analysis data.
The musical result will be more variable than HILITE TRACE. The latter retains the N loudest channels, thus always keeping the most audibly prominent data of the original sound. BLUR SCATTER may or may not pick up audibly prominent data as it makes its random selections. The original sound material will therefore be more variably retained or cast aside, and this variability can be increased by using a time keep and/or a time blocksize breakpoint file.
SEE ALSO: TRACE, BLUR SUPPRESS, SUPPRESS (band)
End of BLUR SCATTER
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
Example command line to expand the area of similar windows :
selfsim selfsim in.ana out.ana 2
inanalfile – input (mono) analysis file
outanalfile – output (mono) analysis file
self-similarity-index – the number of similar windows to replace. Value = 1 uses the loudest window to replace the most similar window, then the next loudest window to replace the window most similar to it, and so on, with appropriate overall-loudness scaling. With value = 2, the loudest windows replaces the two most similar windows, and so on. Thus, if window A replaces window B, and window C is most similar to window B, then window A also replaces window C, etc.
This process systematically replaces spectral windows that are less prominent (i.e.,not as loud, no special peaks ...) with spectral windows in the same file that are more prominent in some way and are the most similar to them in spectral envelope . At the same time it scales amplitudes for any overall difference between the two windows.
It is important to observe the effect of higher values for the self-similarity-index. You might describe this program as a feature repeater. On the one hand it is looking for prominent windows and extending their presence in the sound. On the other hand, these prominent windows are also examined for similarity of spectral envelope. Thus the features of this spectral envelope are repeated more often in the resulting sound – hence the 'self-similar' idea. The overall effect is going to depend on the nature of the prominent windows, and one can expect that a fairly steady-state sound will not be much affected by this process. Try it on exciting sounds!
End of SELFSIM
Return to list of SPEC functions at
top of this file
Return to Main Index for
the CDP System.
Return to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
domain-image – this consists of two strings of letters, separated by a hyphen ('-'); the first string is the domain and the second string is the image, e.g., 'abc-abbabcc'.The domain letters represent a group of consecutive infile analysis windows, e.g., 'abcd'.grpsize – the number of analysis windows corresponding to each letter of domain.
The image is any permutation of, or selection from, these domain letters these letters may be omitted or repeated in the image string, e.g., 'aaaaaaaadaaa'.Each letter then represents a group of grpsize windows and the whole group is treated as one unit in the shuffling process.
SHUFFLE shuffles windows in a PVOC analysis data file. This is a spectral version of 'brassage' (see MODIFY BRASSAGE and MODIFY SAUSAGE). As with most of these programs, the degree of audible difference is going to depend on how much the sound changes in the first place. Shuffling spectral windows in a fairly uniform sound may not have much effect. Experimenting with this program ought to lead to the unexpected.
To illustrate how this works, suppose there is a domain of three letters: A B C. This means that there will be three consecutive windows, numbered 1 2 3 in the order in which they occur in the analysis file.
If the image is ordered C B A, then these three windows will be rearranged in the output into the order 3 2 1.
grpsize has not been set in this example, so the default of 1 is operating. If we then set grpsize to, say 3, the A will stand for 3 windows, starting with 1 2 3, B for windows 4 5 6, and C for windows 7 8 9. When the shuffling takes place according to the Image CBA, the result will be 7 8 9, 4 5 6, 1 2 3. The process continues in this way throughout the analysis file. Larger values for grpsize (e.g., 20) can lead to the repetition of fragments of sound material. The following table summarises these inputs.
BLUR SHUFFLE EXAMPLE Domain Image Groupsize ABC CBA 3 123 456 789 789 456 123 The command line for the above would be:
blur shuffle infile outfile ABC CBA 3
Note that the size of the domain and the value for grpsize are limited by the amount of RAM available on your computer.
The domain-image parameter is what makes SHUFFLE particularly useful. The domain identifies a series of consecutive windows, and the image defines the order of those windows (some may be omitted). Then this whole pattern repeats as a block, affecting each consecutive set of windows in the source file. The control over re-ordering makes it possible to jumble, retrograde, mix forwards and backwards etc. the successive windows. An increasing grpsize makes more of the source audible at any given time, which means that one can play with the degree to which the source can be identified.
Thus one is stepping through the source in chunks of varying size and re-ordering the contents of those chunks. The re-orderings play with familiarity, create repeatable sub-patterns, and will lead to additional timbral changes depending on the degree to which differing spectral data get mixed together.
End of BLUR SHUFFLE
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
-fN extract formant envelope linear frequency-wise, using 1 point for every N equally-spaced frequency channels
-pN extract formant envelope linear pitch-wise, using N equally-spaced pitch bands per octave
-i quicksearch for formants (less accurate)
-sspread degree of spreading of spectrum (Range 0-1; Default is 1).
spread may vary over time
This process introduces noise into the spectrum in a way which is coherent with (i.e., related to) the spectral envelope. The spectral envelope (formants) in each window is retained, and the level in every channel is made to approximate this average spectral contour to a greater or lesser extent, depending on spread. This process tends to exaggerate the less prominent (noise) constituents of the spectrum.
This is a distortion technique which can be handled in a very sensitive manner because spread can be so finely adjusted, or altered in a time-varying manner. If applied to speech, the noise introduces a certain edginess, so if an effect such as unvoiced speech is sought, FORMANTS VOCODE is recommended, which allows some low or high frequency filtering.
End of BLUR SPREAD
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
N – the number of spectral components to reject
N may vary over time
This process is the opposite of HILITE TRACE, which retains the N loudest partials. Whereas with TRACE many components have to be removed before the sound starts to be seriously affected, here the change to infile will be more immediate when a relatively small value for N is used.
The loudest partials will not necessarily relate to either the higher or lower parts of the frequency spectrum. They will relate simply to whatever is most audible prominent in the sound, such as possibly the fundamental of a pitched sound. This will have the effect of revealing the more strictly timbral aspects of the sound.
This process can also be useful in preparing a sound for another process which will benefit from suppressing its more audible features. As an example of this, ... (I did a sound in which I couldn't get rid of the fundamental, which was reinforced by the process, when I really wanted to bring out other things.)
SEE ALSO: TRACE, BLTR, SUPPRESS (band)
End of BLUR SUPPRESS
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
inanal – input analysis file
outanal – output analysis file
timeslots – datafile of (from-to) time-pairs between which bands are suppressed.
lofrq – low frequency limit of band where partials to be suppressed. (Range: 5 to 22050 Hz)
hifrq – high frequency limit of band where partials to be suppressed.(Range: 5 to 22050 Hz)
chancnt – number of most prominent analysis channels in band to be suppressed. (Range: 1 to no. of channels, e.g. 513)
SUPPRESS PARTIALS suppresses the most prominent partials in a specified frequency band, at given times. Lofrq and hifrq set the lower and upper limits of the frequency band, while chancnt gives the number of spectral channels in that band that you wish to suppress. The timeslots datafile gives the times that these are to be suppressed. For the whole file, enter 0 and its approx. end time.
Care must be taken in setting the number of channels (chancnt) correctly to avoid a CDP Error. Divide the frequency band by the channel bandwidth (for the PVOC resolution in use) to give the number of channels in the band.
Suppressing the normally weaker upper partials will make very little difference to the sound. The greatest timbral change is through selective high-pass filtering, that is, in eliminating some of the lower partials, under 1000Hz.
SUPPRESS PARTIALS expands the provision of the earlier BLUR SUPPRESS, which suppresses the most prominent N partials, channel by channel.
...
SEE ALSO: HILITE TRACE, HILITE BLTR, BLUR SCATTER, BLUR SUPPRESS,
End of SUPPRESS PARTIALS
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile – input analysis file made with PVOC
outfile – output analysis file
weavfile contains a list of integers which define successive steps (in windows) through the input file. The start window is always numbered '0'. Steps may be forward or backwards, adding the positive or negative integer of the weave formula to the number of the window at which one has arrived to show which will be the next window selected (see tables below). The step sequence is repeated until the end of the infile is reached.
The weave must obey the following rules:
RULE 1: NO step can exceed 127 forwards or -128 backwards
RULE 2: NO window reached in a weave can be BEFORE the start window of the weave.
RULE 3: FINAL window must be AFTER the weave start window.
Otherwise, weave may be forward or backward, or remain at same window.
BLUR WEAVE is similar to SHUFFLE but is a smoother process which plots a regular course through the analysis windows, weaving backwards and forwards as instructed by the weavfile. Note that the weave process enables one to jump to points forwards or backwards in the soundfile, at which point it takes a single window. Thus one of the keys to understanding WEAVE is to realise that it only takes one window at a time after it has moved somewhere. It doesn't fill in with the windows inbetween. The weavfile therefore creates a sequence of single windows taken from various points relative to the start window. This is what makes it a form of brassage in the spectral dimension. The pattern in the weavfile then repeats, starting from the window it has now reached.
BLUR WEAVE can shorten or lengthen a sound:
Shorten: suppose 3 windows are selected in a way which spans a total of 7 windows. For example, the weave pattern 3, -1, 5, starting from window 0 will take 3 steps, thus moving to window 3 (and selecting it for use in the outanalysisfile), move back one (3-1=2) and take window 2, and move forward 5 (2+5=7) and take window 7. Thus only 3 out of the 7 possible windows are written to the outfile, making the latter shorter by a ratio of 3:7.
Lengthen: suppose a few windows are used over and over (the hovering effect mentioned above), such as with the step pattern 2 -1 1 -1 1 -1 1 -1 1 2. Here 3 windows are expanded to 10, a ratio of 10:3.
weavfile produces reduction by 0.43 new window count 0 1 2 3 source window 0 1 2 -1 3 3 4 5 6 7 5
weavfile produces expansion by 3.33 new window count 0 1 2 3 4 5 6 7 8 9 10 source window 0 1 -1 -1 -1 -1 2 2 1 1 1 1 3 4 2
WEAVE can be used to shorten or extend a sound and/or to blur its characteristics: by taking windows out of sequence, earlier (backwards) or later (forwards) in the source. Windows taken from relatively distant locations of course mix up the sound quite a bit. A very different effect is achieved by having the windows hover around a certain area. The fact that the pattern repeats leads to additional possibilities to create pulsations in the sound, though the timbral effect is likely to vary due to changing spectral content as the weave moves through the source.
End of BLUR WEAVE
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
These various functions to shuffle windows provide a tremendously varied set of tools with which to re-pattern the stream of analysis data. Before comparing them to one another, let's take a moment to revise just what a window is and just what of the original sound it encapsulates.
A window contains the spectral data in all the channels for a given duration of infile. If the number of samples used for the window is 1024 (indicated by the -cpoints flag – formerly the -N flag), then the number of channels is 1024 divided by 2 (= 512) plus an extra channel (= 513).
This number divided by the sample rate gives the duration of one window. E.g., 512/44100 = 0.01163 sec. The hamming window usually used envelopes these samples in a certain way to maximise amplitude but avoid clicks by starting from and returning to 0. (The windows are made to overlap in order to avoid holes created by the enveloping.) Therefore, shuffling windows means moving about these little chunks of time, with all the timbral data – the changing frequency/amplitude content contained therein. The new juxtapositions of this timbral data are likely to lead to new colours in the output sound.
A brief overview of the SHUFFLE functions can help to clarify how they differ, and therefore what one may expect to gain by using each of them.
End of Technical Discussion
Return to List of BLUR functions
at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index