Names in brackets mean that these are separate programs. The others are sub-modules of FILTER.
When filters of time-varying Q are used, the output level tends to drop as the Q increases. This can be compensated for later by using MODIFY LOUDNESS, Mode 5 (Balance sources), submitting the original file and the filtered file as the two sources. The resultant soundfile will have the sound of the filtered file with the amplitude contour of the original sound.
1 HARMONIC SERIES over lofrq
2 ALTERNATE HARMONICS over lofrq
3 SUBHARMONIC SERIES below hifrq
4 HARMONIC SERIES WITH LINEAR OFFSET: param = offset in Hz
5 EQUAL INTERVALS BETWEEN lofrq and hifrq: param = number of filters
6 EQUAL INTERVALS BETWEEN lofrq and hifrq: param = number of semitones in the interval
infile – input soundfile to filter
outfile – output (filtered) soundfile
Q – 'tightness' of filters (Range: 0.00100 <= Q < 10000) – the higher the value the more tightly the filter is focused on the centre frequency of that filter.Q may vary over time.gain – overall gain (Range: 0.00100 to 10000.0)
lofrq – low frequency limit of filters (Range: 10 to sample_rate/3)
hifrq – high frequency limit of filters (Range: lofrq+ to sample_rate/3)
param –-sscat – random scatter of filter frequencies (Range: 0 to 1; the Default is 0)
- in Mode 4, offset in Hz
- in Mode 5, number of filters
- in Mode 6, number of semitones in interval
-d – double filtering
It may be helpful to compare this function with FILTER USERBANK and FILTER VARIBANK, which allow the user to specify the frequency centres for the bank of filters. FILTER BANK handles the definition of the frequencies for you, by means of a variety of presets.
Each Mode operates a different preset. This is how they work:
- Mode 1 – The harmonic series is formed by multiplying a given frequency by an ascending series of integers: 1, 2, 3, 4, etc. The harmonic relationships produced are sonorous, but depart from the structure of triadic chords higher in the series. This Mode builds this type of harmonic relationship onto lofrq, which may be the actual pitch of a pitched tone or any arbitrary pitch. Because this is being done with a filter process, other sonic material is cleaned away, so a cleaner (and possibly quieter) sound may result, with more complex noise elements removed. Quite a bit of gain may be required – values of 40 or 50+ in this situation will not be unusual. Set hifrq as high as possible less too few harmonics are produced. This clean sound may provide a good input for spectral time-stretching, for example, to achieve a well-tuned sound with a minimum of artefacts.
- Mode 2 – This will be like Mode 1 but with every other harmonic omitted: the odd numbered partials are retained. The resulting sound will probably sound more 'hollow'.
- Mode 3 – The 'subharmonic series' is the intervallic inverse of the harmonic series. Thus the departure from triadic-type intervals increases as the series descends. The resulting sound has a deeper tone and is somewhat hollow and somewhat inharmonic.
- Mode 4 – The offset in Hz is added to each harmonic. This will displace them so that they will no longer be exact multiples of the fundamental. This means that the cycles of the waveforms no longer line up at nodes, which introduces an 'inharmonic' dimension into the sound, heard as an increase in timbral components.
- Mode 5 – Here we depart from the harmonic series and simply divide up the specified frequency range into an equal number filters: how many is specified by the user with param. The first effect this equal spacing will have is to create inharmonic (rather than integer multiple) relationships between the partials. A low number of filters will produce an 'open' sound, and a high number of filters will produce a denser, richer sound. These might be very interesting sounds to timestretch.
- Mode 6 – By specifying the interval (in semitones), we in effect repeat these intervals within the frequency space between lofrq and hifrq, thus producing complex chords composed of the same interval 'piled up'. This is a quick way to 'harmonise' a sound, with the resulting density dependent on the size of the interval. It is well worth re-running this function, entering intervals from 1 to 7, for example, to hear what kind of transformations will be produced. Some unexpected resonances may result; they could be filtered out later.
The use of the presets provide quick ways to create a variety of harmonic and inharmonic transformations. The resulting sounds have a sonorous, chordal quality. One may need to push the gain control as far as it can go without overloading, less too much signal is lost through the filtering process.
The resulting sounds can be used for rich, sonorous textures or clean time-stretching, or other spectral transformations.
Having said that, the 'chordal quality' will be diminished if the Q is lower (i.e., less pitch focus). Also note that a lower Q 'lets through' more of the original signal, so the gain will also have to be reduced. For example, with a Q of 100, a gain of 100 worked fine; but with a Q of 100, a gain of 5 was required (for the source sound used).
End of FILTER BANK
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 HARMONIC SERIES over lofrq
2 ALTERNATE HARMONICS over lofrq
3 SUBHARMONIC SERIES below hifrq
4 HARMONIC SERIES WITH LINEAR OFFSET: param = offset in Hz
5 EQUAL INTERVALS BETWEEN lofrq and hifrq: param = number of filters
6 EQUAL INTERVALS BETWEEN lofrq and hifrq: param = number of semitones in the interval
infile – input soundfile to filter
outtextfile – output textfile containing frequency data
lofrq – low frequency limit of filters (Range: 10 to sample_rate/3)
hifrq – high frequency limit of filters (Range: lofrq+ to sample_rate/3)
NB: The sample rate of the input soundfile determines the filter frequency range.param –-sscat – random scatter of filter frequencies (Range: 0 to 1; the Default is 0)
- in Mode 4, offset in Hz
- in Mode 5, number of filters
- in Mode 6, number of semitones in interval
This function carries out the same operations as FILTER BANK, but writes the resulting frequencies to a textfile rather than applying them to a soundfile.
NB: An existing outtextfile of the same name will be overwritten without checking with you first.
See FILTER BANK for the meaning of each Mode.
This textfile can then be edited and is then available as an input to FILTER USERBANK and FILTER VARIBANK, the filter bank functions which take a user-defined set of filter centre-frequencies as an input.
End of FILTER BANKFRQS
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
Example command line to create a randomised filterbank:
filtrage filtrage 1 filtout.txt 10 4 36 72 1 0.75 0.5 0.5 0 2 0
1 Generate a fixed filter data file
2 Generate a time-varying filter data file
outfiltdata – output text file containing the time-varying filterbank data for FILTER VARIBANK
dur – the duration spanned by the output filter file
cnt – the number of parallel filters
min – the minimum MIDI value for the filters
max – the maximum MIDI value for the filters
distrib – the distribution of pitch values:
• 1 = pitch-linear
• > 1 squeezes the filter towards the low pitches
• < 1 squeezes the filter towards the high pitches
rand – randomisation of the filter pitches
ampmin – the minimum filter amplitude. The maximum amplitude allowed here is = 1.0
amprand – randomisation of the filter amplitudes
ampdistrib – the distribution of the filter amplitudes:
• 1 = increasing with pitch
• -1 = decreasing with pitch
• 0 = random
• Intermediate values give decreasing degrees of randomisation.
timestep – the amount of time in seconds between each specified set of filter-pitches
timerand – randomisation of the timestep
-sseed – non-zero values fix the randomisation so that, on repeating the process identical random values are produced.
Filtrage produces a filter data file for FILTER VARIBANK, with random frequencies. These are distributed within the desired frequency range either evenly, or skewed toward the top or bottom of the range. The command line above (4 filters, MIDI range 36-72, and a set of values every 2") produced this file:Time Frequencies 0.00 36.652402 0.962668 48.366554 0.835011 54.429201 0.673751 60.270501 0.514595 2.00 37.025983 0.993088 46.420885 0.844629 54.091361 0.700716 62.949427 0.506138 4.00 37.794362 0.842264 48.314848 0.675147 54.824309 0.972373 61.072253 0.506926 6.00 37.754604 0.961428 46.912400 0.697998 56.042592 0.852223 61.980918 0.539829 8.00 38.631340 0.983648 47.445116 0.874975 55.960192 0.708066 60.152669 0.525479 10.00 38.285569 0.868233 45.837905 0.707601 56.749790 0.988629 62.810274 0.511495The randomisation is not as great as you might imagine, but permits a controllable degree of variation of amplitude, frequency and output times, which can be further adjusted manually as required.
...
End of FILTRAGE
Return to list of FILTER functions at
top of this file
Return to Main Index for
the CDP System.
Return
to 'Groucho' Time-Domain Index
1 Boost or cut below a given frequency
2 Boost or cut above a given frequency
3 Boost or cut a band centered on a given frequency
infile – input soundfile to filter
outfile – output (filtered) soundfile
bandwidth – filter bandwidth in Hz – Mode 3 only
boost/cut – amplitude boost or cut, in dB
frequency – the frequency of the filter in Hz
-sprescale – scales gain on the input to the filter
This filter is referred to as 'fixed' because a single frequency of fixed amplitude roll-off (Q), is preset within the function.
FILTER FIXED requires very precise frequency inputs. It can therefore be useful to analyse the sound and use SPECINFO PEAK get a profile of the frequency bands in which most energy occurs.
For example, a trombone sound playing an F below Middle-C (174.61 Hz) showed almost no change when this frequency was used as the input to FILTER FIXED. SPECINFO PEAK showed that the energy was concentrated between 640 Hz and 905 Hz. Mode 1 and Mode 2 inputs relating to these frequencies worked fine: e.g., filtering out below 905 Hz or above 640 Hz really made a difference; similarly, filtering out a 200 Hz band centered on 700 Hz.
FILTER FIXED can be a quick way of achieving three common filtering operations: hi-pass (Mode 1: cut below), lo-pass (Mode 2: cut above), or notch (Mode 3: cut around), given a precise knowledge of the frequency area to be filtered and an acceptance of a fixed, average amount of Q.
End of FILTER FIXED
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 Enter filter pitches as frequency, in Hz
2 Enter filter pitches as MIDI note values
infile – input soundfile to filter
outfile – output (filtered) soundfile
datafile – contains the pitch and amplitude of the filters (paired, one pair on each line)Q – 'tightness' of filters (Range: 0.00100 <= Q < 10000) – the higher the value the more tightly the filter is focused on the centre frequency of that filter.
- Pitch: as Hz or MIDI note values (Range: 10Hz to sample_rate/3)
- Amplitude: amplitude level (Range: 0.00100 to 10000.0 or entered as dB values)
- Remember that it is possible to create a file of frequency data with FILTER BANKFRQS and then edit it to add the amplitude information. Such files can be useful because of their harmonic structure.
Q may vary over time.gain – overall gain (Range: 0.00100 to 10000.0)
delay – average delay in seconds between iterations (Range: > 0 to 32767)
dur – (min) duration of output file (must be longer than infile
-sprescale – scales gain on the input to the filtering process (Range: 0.0 to 1.0; the Default is 1.0)If set to 0.0, FILTER ITERATE automatically divides the input level by the maximum number of sound overlays occurring in the iteration process.-rrand – randomisation of the delay time (Range: 0 [none] to 1 [max])
-ppshift – maxmimum pitch shift of any segment in (fractions of) semitones (Range: >= 0)
-aashift – maximum amplitude reduction of any segment (Range: 0.0 to 1.0)
-d – double filtering
-i – turn off interpolation during filtering (makes it fast, but dirty)
-e – add exponential decay: each segment gets quieter before the next segment enters
-n – turn off normalisation: segments may grow or fall in amplitude quickly; gain settings will have more impact when -n is used – but use this cautiously, as overload may occur, which FILTER ITERATED does not report.
Note that the duration of the output soundfile is set with dur, it will usually be considerably longer than infile to provide space in which the iterations can occur. In effect, this determines the number of iterations there will be, given the value for delay, the time between iterations.
If you receive a message about 'insufficient memory', you can increase the buffer capacity by increasing the buffer size. This is done by setting the environment variable CDP_MEMORY_BBSIZE. This is 1 megabyte by default. The units are 1K each, so a buffer size of e.g., 6 megabytes can be set with the phrase: 'set CDP_MEMORY_BBSIZE=6000'. Note that there must NOT be spaces before and after the equals sign. This can be done in the existing DOS window, or in autoexec.bat so that it sets this size upon boot-up. If the latter, the new size will not come into force until you re-boot.
More iterations means that the sound will be more filtered by the time it reaches the end. This has an important relationship to Q, because the higher the Q, the more the original sound material disappears into a pure and clean resonance as the sound progresses. Similarly, this resonance can be made to modulate nicely by using some pitch transposition with pshift, producing a weaving of differently tuned strands.
Delay is perhaps the key parameter, because it determines the degree of overlap with each overlap – or no overlap if it is longer than infile. A long dur (many iterations) with a delay longer than infile means a series of differently filtered repeats, with gaps between them. Higher values of rand, will close some of these gaps by introducing random overlaps. However, with no or occasional overlap, there will be no (or very little) resonance effect.
NB: Each iteration begins at the beginning of infile and plays delay amount of infile, before the next iteration commences over the top of it. How your source sound begins will greatly affect the results, especially the sharpness of attack and initial amplitude. Normally, it should have a high amplitude in order to maximise signal as the filtering progresses.
A soundfile with a strong attack and a short delay (e.g., 0.2 or 0.1 sec) will produce a series of pulsations which gradually disappear into resonance, depending on Q. It takes quite a low Q (e.g., well under 10) to hold off the resonance effect for a bit. The timing of the transition from the source sound to a resonant effect is handled by balancing dur, delay and Q.
FILTER ITERATED is processor intensive. Very short values for delay will increase the processing time, though this does not appear to affect the demands on available RAM.
The results achieved with FILTER ITERATED can vary widely. Here is a listing of some of the main effects:
- long dur – many iterations and plenty of time for the repeated filtering to alter the sound
- very short delay (< 0.05) – a granular sound which gradually becomes a rapidly pulsating resonance; add some pshift and it may become a soft, modulating wash
- short delay (e.g., ca 0.1 or 0.2) – rapidfire pulsations
- the use of exponential decay (-e) will increase the sense of pulsation by dipping the amplitude between iterations
- long delay – gross iterations of the beginning of the sound, becoming a bit softer as the filtering takes effect
- delay longer than infile – delay amounts of infile with gaps between each iteration; add some rand factor and there will be some random overlapping of segments
- high Q – the tendency to dissolve into resonance will be considerably greater
- high Q with pshift – the resonance produced as the sound progresses will tend to warble and weave tuned strands
End of FILTER ITERATED
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 Pass-band and stop-band are given as frequency in Hz
2 Pass-band and stop-band are given as (possibly fractional) MIDI note values
infile – input soundfile to filter-sprescale – apply gain on the input to the filtering process – avoid overflows (Range: 0.005 to 200.0)
outfile – output (filtered) soundfile
attenuation – gain reduction of the filter, in dB (Range: 0 to -96)The greater the attenuation, the sharper the filter – but it takes longer to calculate.pass-band – last pitch to be passed by the filter
stop-band – first pitch to be stopped by the filter
- If stop-band is above pass-band, this is a low pass filter.
- If stop-band is below pass-band, this is a high pass filter.
The lo-pass filter 'lets through' all of the sound below pass-band and attenuates (makes gradually softer) the frequencies higher than pass-band, ending at stop-band.
The high-pass filter 'lets through' all of the sound above pass-band and attenuates (makes gradually softer) the frequencies lower than pass-band, ending at stop-band.
The key is how the pass-band and stop-band are placed against the significant frequencies of the source sound. As mentioned elsewhere, it may be useful to perform a spectral analysis on infile and make a report with SPECINFO PEAK.
An interesting fact about the operation of the filters has recently been observed. With the hipass-lopass filter (FILTER LOHI – and possibly with other filters?), if your source file has LOTS of silence at the end, the filter suddenly slows down by a factor of 20 (or more!!) in the silent buffers. At the moment, we don't know why this happens – there's no known bug in the code. So you are recommended to 'topntail' (ENVEL DOVETAIL) any long sounds that fade away to nothing before running them through filters.
These basic filter operations can be used to clean up a sound by removing excessive bass with a high-pass filter, or some hiss (if it is above significant frequencies in the source) with a low-pass filter.
Musically, a high-pass filter makes the sound 'thin' by removing the bottom, and a low-pass filter makes the sound deep and rumbling by removing the top.
End of FILTER LOHI
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 Allpass filter (phase-shifted)
2 Phasing effect
infile – input soundfile to phase shift
outfile – output (phase-shifted) soundfile
gain – amplitude adjustment (Range: -1.0 to 1.0)In Mode 2 the phasing effect increases as gain increases from -1.0, but a gain of 1.0 will produce complete phase cancellation and the output signal will be 0.delay – time in milliseconds between return of delayed material (Range: 0.045 to 1644.35 ms)Delay may vary over time.-sprescale – scale gain on the input to the filtering process (Range: 0.0 to 1.0; the Default is 1.0)
-l – linear interpolation of changing delay values (Default: logarithmic)
The 'phasing effect' is a kind of sweeping band passing through the sound, such as is sometimes heard when aeroplanes fly overhead, and is much used in popular music.
Increasing the gain factor has some effect on the reverberant quality of the sound. So does increasing the delay time. The sound will still sound quite dry with delay < 20 (ms). Between about 20 and 45 ms there is a touch of resonance in the sound. Around 50 ms there is significant echoey reverberation (though somewhat granulated), and after 100 ms we start to hear larger portions of the sound repeating.
Overall, the degree of reverberant effect is controlled by increasing both the gain and delay parameters in tandem.
Used with care, this function can be used to produce a variety of reverberant effects. The mid-range results are similar to the sound of a loud clap in a stairwell or under the overhang of a cement building.
Longer delay times will produce echo effects, but also see MODIFY REVECHO for echoes proper, as well as the other REVERB options.
End of FILTER PHASING
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
infile – input soundfile (MONO).
outfile – output soundfile (MONO, STEREO or MULTI-CHANNEL).
streams – number of output streams that phase-interact. (Range 2 to 8; >= ochans)
phasfrq – frequency of packets (phase shifts forward then back in a single packet). (Hz)
shift – maximum phaseshift with packet. (Range 0 - 12 semitones)
ochans – number of output channels (not greater than the number of streams). (Range: 1 to 8)
-ooffset – streams may be time-offset from one another: offset is the time-offset of the most-offset stream. (Range 0 - 500 mS)
The other streams are offset by intermediate amounts.
-s – output channels (loudspeakers) encircle the audience; use only with more than two output channels
-e – prints warnings about rounding-errors in calculating time-shifted streams.
shift and phasfrq can vary through time, but extreme phasfrq changes or reversals will produce anomalous output.
PHASOR produces a phasing effect, with mono, stereo or multi-channel output. The signal is divided into a number of output streams that phase-interact. The key paramaters are phase frequency (phasfrq) and pitch shift shift, which actually drives the phasing process (value 0 has no phasing effect). There is also an optional offset parameter which time-offsets the streams: even without pitch shifting, this is an effective way of creating a pseudo-stereo effect out of a mono signal, since it is phase and timing differences that create the stereo illusion.
The program can produce a chorus effect, too, by setting phasfrq fairly low (e.g. 0.5), plus some pitch shift (e.g. 2) and offset (>40mS). As a phasing effect, PHASOR surpasses the earlier FILTER PHASING 2, although the latter has some value for chorusing.
...
End of PHASOR
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 High-pass
2 Low-pass
3 Band-pass
4 Notch (band-reject)
infile – input soundfile
outfile – output soundfile
acuity – tightness of the filter (Range: 0.000100 to 1.0)Smaller values give a tighter filter.gain – overall gain on output (Range: 0.001000 to 10000.0)Rule of thumb: If acuity = (1/3)-to-power-n, gain = (2/3)-to-power-n.lofrq – lowest frequency to sweep to (Range: 10.0 to sample_rate/2)
hifrq – highest frequency to sweep to (Range: 10.0 to sample_rate/2)
sweepfrq – frequency of the sweep itself (Range: 0.0 to 200; the Default is infile_duration/2)For example, to sweep once over the time of the soundfile, set sweepfrq to infile_duration/2 and set phase to 0 (upsweep) or 0.5 (downsweep)-pphase – start position of the sweep (Range: 0 to 1; the Default is 0.25 – midway along the rising curve)Imagine a series of points along a rising and falling curve. The curve begins at 0 (= lofrq) rises to 0.5 (= hifrq), and moves on to 1, falling back to lofrq. The phase value identifies where along this curve the sweep shape begins.Acuity, lofrq, hifrq and sweepfrq may all vary over time.
This filter is most effective in Band-pass or Low-pass mode, with low acuity. It is difficult to get a result in High-pass mode. A tight filter makes the sweep more audible, but too tight and it may appear as a thin sine wave (resonance) which doesn't really connect with the sound.
A fairly long soundfile and a fairly broad range of frequencies gives this function something to work with.
Watch for overflows and reduce the gain as necessary.
With acuity less than 0.1 (tight) and a sweepfrq less than 0.5 (slow), one can produce wah-wah effects. Sometimes it may be useful to cut only a segment of these timbrally modulating glissandi for use as a separate soundfile.
With sweep frequencies of, for example, 20 or 100, and perhaps a wider acuity, one can produce fluttering effects, the wider the acuity, the 'looser' the flutter/flapping effect.
Sometimes, with tighter acuity settings, one can hear movement through the harmonic overtone series.
End of FILTER SWEEPING
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 The pitches to filter are entered as frequency in Hz
2 The pitches to filter are entered as MIDI note values
infile – input soundfile to filter
outfile – output (filtered) soundfile
datafile – contains the pitch and amplitude of the filters (paired, one pair on each line)Q – tightness of filter (Range: 0.001000 to 10000.0) – the higher the value the more tightly the filter is focused on the centre frequency of that filter.
- Pitch: as Hz or MIDI note values (Range: 10Hz to sample_rate/3)
- Amplitude: amplitude level (Range: 0.00100 to 10000.0 or entered as dB values)
- Comment lines (starting with ';') may be used.
Q may vary over time.gain – overall gain (Range: 0.001000 to 10000.0)
-d – double filtering
Here Q may vary over time, but the frequency settings are fixed for the duration of the sound. The tightness of the Q determines how much of the original sound 'comes through', or, to put it another way, the degree to which the sound is tuned to the frequencies defined in the filter bank.
Much of the effectiveness of the function therefore depends on the design of the datafile. Normally, dB amplitude settings will fall within a 0dB to -96dB range, but the function does allow dB greater than 0 (i.e., applying gain to the sound). If you have existing filter bank files with dB greater than 0 and they overload, you can use gain with values below zero to scale the amplitudes back within range – i.e., without having to edit your file.
As with any graphic equaliser, this function can be used to emphasize or de-emphasize different frequency regions of the source sound.
It can also be used to impart a harmonic sonority to a sound. The higher the Q, the clearer the pitched effect (e.g., values between 300 and 1000). Given a source with a rapidly changing amplitude envelope, this may come across as a series of harmonic scintillations, whereas a sound with a steadier amplitude envelope will become a chord.
End of FILTER USERBANK
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 High-pass
2 Low-pass
3 Band-pass
4 Notch (Band reject)
infile – input soundfile to filter
outfile – output (filtered) soundfile
acuity – tightness of the filter (Range: 0.000100 to 1.0)Smaller values give a tighter filter.gain – overall gain on output (Range: 0.001000 to 10000.0)Rule of thumb: If acuity = (1/3) -to-power-n, gain = (2/3)-to-power-n.frq – frequency of the filter (Range: 10.0 to sample_rate/2)Acuity and frq may vary over time.
The acuity parameter controls the rate at which the amplitude decreases as one moves away from frq. The faster the rate, or, to put it more technically, the steeper the slope of the roll-off, the more the resulting sound is focused on frq (Band-pass), or the area above (Hi-pass) or below (Lo-pass) frq, or omits the area around frq (Notch). It is best not to make the acuity too tight with this filter, as the resonance produced may become unpredictable.
Hi-pass and Lo-pass are not very effective in FILTER VARIABLE, which has been designed as a 'lightweight', fast filter. For more powerful Hi-pass and Lo-pass filtering, it is recommended that you use FILTER LOHI.
Because a filter often creates resonance (boosts the amplitude of frequencies in a certain frequency area), especially with a broader roll-off region, overload can easily occur. FILTER VARIABLE reports any overflows, in which case the outfile should be deleted without being played and the function run again with a reduction in the gain.
FILTER VARIABLE provides a convenient environment for exploring the effect of a variety of filter types on a given sound.
End of FILTER VARIABLE
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
1 The pitches to filter are entered as frequency in Hz
2 The pitches to filter are entered as MIDI note values
infile – input soundfile to filter
outfile – output (filtered) soundfile
datafile – contains lines of data for filter bands at successive timesQ – tightness of filter (Range: 0.001000 to 10000.0) – the higher the value the more tightly the filter is focused on the centre frequency of that filter.
- Each line contains data in a certain format.
- This format is: Time: Pitch1 Amp1 [Pitch2 Amp2 etc...]
- The Pitch and Amplitude values must be paired, but any number of pairs can be used on one line.
- All the lines in datafile must have the same number of pairs.
- To eliminate a band in any line(s), set its amplitude to 0.0)
- Time values are in seconds and must be in ascending order and >= 0.0
- Pitch values may be entered as frequency in Hz (Mode 1) or as MIDI note values (Mode 2)
- Amplitude values may be numeric or expressed as dB values (e.g., -4.1dB)
- Comment lines may be used – they must begin with ';'
- In FILTER VARIBANK2, you can also specify partials and their relative weighting. To do this you add a '#' sign on the next line at the bottom of the datafile, followed by lines which specify the way the partials vary over time. The default (in VFILTERS) is harmonics at 2,3,4,5,etc. times the given frequency (as MIDI pitch or frequency). In FILTER VARIBANK2 you can specify the partials in this format to add changing timbral colouring (also see below):
# Time Partials with Amplitude weighting (pairs)] 0 1 .5 2.2 .7 4.6532 .5 4.5 1 .25 2.4 .2 3 .14- See a synopsis of the file format in CDP Files & Codes.
Q may vary over time.gain – overall gain on output (Range: 0.001000 to 10000.0)
-hhcount – number of harmonics of each pitch to use (Default: 1)The high harmonics of high pitches may go beyond the Nyquist frequency (sample_rate/2)-rrolloff – decrease in amplitude level in dB from one harmonic to the next (Range: 0 to -96)
-ttail – duration of decay tail (Range: 0 to 20 secs)
-d – double filtering
The frequencies specified in the datafile are in fact the centre-frequencies of a filter band. With a tight Q, these specific frequencies will be heard as pitches; with a relaxed Q, these specific frequencies will locate a relatively fuzzy pitch region. The pitch will glissando between different pitch levels when these differ between time points.
Note that if the datafile has, for example, 10 Pitch/Amplitude pairs on a line, each column of paired values relates to a given 'band'. Thus the change over time for each band will be specified by the sequence of paired values in a given column. The value for Q is not contained in the datafile, but the time points for changes in Q can be made to match or vary from those in the datafile.
In the additional lines supplied for the FILTER VARIBANK2 option, the first value in each line is a time. The following values are in pairs representing the partials (they can in fact be ANY number) and their relative amplitude. Note that as with the frequency (or MIDI) data already given, there must be the same number of entries in each line. If you want to omit a partial at given times, give it an amplitude of 0.
Thus you might omit the second partial during specified times:
Time Partials with Amplitude weighting (pairs)] 0 1 0.3 2.237 0.4 7.615 0.8 1 1 0.3 2.37 0 7.615 0.8 [partial 2.37 disappears 2 1 0.3 2.37 0 7.615 0.8 during these times] 3 1 0.3 2.37 0.4 7.615 0.8The partials could also be changing:0 1 0.3 2.237 0.4 7.615 0.8 1 1 0.3 3.34 0 14.22 0.8 2 1 0.3 16.3 0 8.322 0.8 3 1 0.3 22.6 0.4 7.615 0.8Here's a complete example, with VARIBANK + VARIBANK2 data:
Example Data and Q settings for FILTER VARIBANK Datafile | Q-file Time Pitch1 Amp1 Pitch2 Amp2 Pitch3 Amp3 Description | Time Q Description 0.0 53 -3dB 53 -3dB 53 -3dB F below Middle C | 0.0 5000 Very tight focus 2.0 65 0dB 53 0dB 41 0dB F in 3 octaves | 2.0 70 Very fuzzy 3.0 57 -6dB 54 -6dB 50 -6dB D-Major chord | 3.0 2000 Mid-range pitch focus 4.5 67 -3dB 62 -3dB 59 -3dB G-major, 1st inversion | 4.5 2000 No change 7.0 64 -3dB 60 0dB 60 -2dB 2 Middle C's & 1 E | 7.0 5000 Very tight pitch focus # '#' sign for VARIBANK2 information to come Time Par1 Amp2 Par2 Amp3 Par3 Amp1 NB: Same number of pairs on each line 0 1 0.5 2.2 0.7 4.6532 0.5 Time + partial - amplitude_weight pairs 4.5 1 0.25 2.4 0.2 3.0 0.14 Put zero amp if you want to omit a given partial The whole file (version for FILTER VARIBANK2) therefore looks like this:
0.0 53 -3dB 53 -3dB 53 -3dB 2.0 65 0dB 53 0dB 41 0dB 3.0 57 -6dB 54 -6dB 50 -6dB 4.5 67 -3dB 62 -3dB 59 -3dB 7.0 64 -3dB 60 0dB 60 -2dB # 0 1 0.5 2.2 0.7 4.6532 0.5 4.5 1 0.25 2.4 0.2 3.0 0.14Higher values for Q filter the sound more, which reduces overall amplitude. Higher Q therefore allows for and sometimes requires higher gain levels. Similarly, lower values for Q 'let through' more of the original sound, so there is less amplitude reduction through the filtering process, and therefore more likelihood of overflows. Higher values for hcount also increase the tendency to overflow, because there are more harmonics in the filter effect. These harmonics enrich the sound.
The data in the two files used in the above example were used in a command line with Mode 2, a gain of 20 and hcount's of 6 and 12. Also try it without using hcount to hear the difference. This was the full command line:
The input was a sustained trombone tone on F below Middle C, which is why the filter frequency began at MIDI note 53.filter varibank2 2 infile outfile datafile.txt qfile.txt 20 -h12Mapping out the data with a little diagram may be helpful:
- Vertical lines to mark the time points
- A series of horizontal lines rising and falling towards each time point (or not changing) to indicate pitch levels (use MIDI note values & Mode 2 unless exact frequencies are crucial); label the MIDI notes on your diagram.
- Write an amplitude value in dB next to each pitch, possibly in a different colour
- Put 'tight' at the top left of your diagram and 'relaxed' at the bottom left, in a different colour: this is the Q scale
- Draw a line for changing Q in the Q colour, changing direction – or staying the same – at each time point.
The facilities of FILTER VARIBANK are a composer's dream. Not only can the frequencies of the filterbank be specified in precise detail, but the frequencies may move in time. Not only can the tightness of the filters – the degree to which they focus on pitches – be specified, but this focus can also change over time. The number of harmonics and the amplitude 'rolloff' are also under the composer's control.
It is possible, therefore, to play with relationships between pitched tones and complex sounds with some degree of noise components. Working with sounds accepts all kinds of sonic material into the fabric of musical discourse – but pitch need not be excluded simply because a broader range of materials is being used. A tool like FILTER VARIBANK makes it possible gradually to focus complex sonic material into specific pitches, thus creating links with other pitched elements, such as a broader harmonic scheme and the use of acoustic musical instruments.
End of FILTER VARIBANK
Return to list of FILTER functions at top
of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index
inpitchfile – contains a list of MIDI Pitch Values or frequency pitch values, with one or more values on each line
generic_outfilterbankfile – each line is converted into a data file for a pitched pitch(es) filter for FILTER VARIBANK. The outfile names are whatever you put for generic_outfilterbankfile plus a '0', '1', etc. appended, producing a different file for each line in inmidifile.
FILTER VFILTERS creates one or more data file(s) for the FILTER VARIBANK process from a list of MIDI Pitch Values (MPV's), one file for each line of input. The output basic filterbank file is a fixed (not time-varying) pitches filter. If the inmidifile is:
48 63.5 70 74 81the generic_outfilterbankfile will be (edited here for conciseness):[Time MPV Amp MPV Amp MPV Amp MPV Amp MPV Amp] 0 48 1 63.5 1 70 1 74 1 81 1 10000 48 1 63.5 1 70 1 74 1 81 1The second line with time = 10000 means that the filter values given extend (without changing) from 0 seconds to 10000 seconds (i.e., the end of the file). The program adds it because it needs to have two time values in a varibank data file.It actually doesn't matter whether the values in inmidifile are MIDI or frequency values. The input is just treated as numbers which go into the output. If you put in what you think is MIDI, the output will be in MIDI. If you put in what you think is frequency, the output will be frequency. Many musicians find it easier to think of harmonies in the MIDI Pitch Value system, so this is illustrated here. These MIDI Pitch Values, please note, may be fractional (microtonal).
Note that the data in VARIBANK filters can also be generated, and transposed from the Sound Loom Table Editor.
BATCHFILES can be expanded to use a sequence of different values on the same, or a series of different sources, using the facilities in the Sound Loom Table Editor.
FILTER VFILTERS will help you to generate data files for the FILTER VARIBANK process more quickly.
End of FILTER VFILTERS
Return to List of functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to 'Groucho' Time-Domain Index