Return
to Main Index for the CDP System.
Return
to Spectral Index
infile, infile2 – input analysis files made with PVOC
outfile – output analysis file
-iinterp – the degree of replacement (Range: 0 to 1)
interp may vary over time
Formerly SPECCROS, this function makes a weighted substitution of the amplitude of one spectral envelope for that of the other. It is therefore a fairly straightforward way to achieve a crossover between the timbral characteristics of two different sounds.
The spectral envelope contains the data for spectral peaks/formants. Thus the transfer of timbral characteristics. However, it is not a full substitution of the whole envelope, but only a (weighted) crossover. It can therefore be used, for example, to prepare two sounds for morphing, especially when the timbral characteristics differ to a considerable extent. By moving one sound gradually into the timbral field of the other prior to morphing, a smoother morph transition can be achieved.
End of COMBINE CROSS
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile, infile2 – input analysis files made with PVOC
outfile – output analysis file
-ccrossover – the amount of the 2nd spectrum subtracted from the 1st (Range: 0 to 1)
crossover may vary over time-a retain any subzero amplitudes produced (Default: set these to zero)
In COMBINE DIFF one spectrum is subtracted from another on a channel-by- channel basis (weighted by crossover). If amplitude in any channel goes negative, you can opt whether or not to retain the negative amplitudes.
By default, the setting of any amplitudes < 0 to 0 protects against negative overflow, but you can use the -a flag and see what happens.
This is a way to change a sound on a very experimental basis.
Also see: COMBINE SUM.
End of COMBINE DIFF
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile(s) – an number of input files made with PVOC (at least two), which will be processed sequentially in the same order as given on the command line
outfile – output analysis file
leafsize – number of analysis windows in each leaf (group of windows)
COMBINE INTERLEAVE interleaves N spectra from N analysis files, thereby constructing a new analysis file as output. The spectral characteristics of several sounds are thus amalgamated into one new sound.
The degree to which the original sounds are retained depends on the number of analysis windows used.
A leafsize is simply a specified group of analysis windows. If leafsize is 1, then the windows from each file are interleaved one at a time. If leafsize is 2, then the windows from each file are interleaved two at a time, etc. Larger leafsize retains more recognisable chunks of the original sounds.
With very small values for leafsize, the various sounds are mingled quite thoroughly and relatively smoothly – relative to the nature of the input sounds, though a somewhat grainy effect may result. With larger values for leafsize, say 8, 10, 20 or more, more of each component sound is heard in each leaf and the result is more like a churning or pulsating effect as the new soundfile oscillates among the sound material in the various source soundfiles. To churn means to shake, stir or agitate violently, as when making butter from milk, which is nicely descriptive of what happens when several sound sources, each fairly strong, are interleaved in this way. [It's one of my favourites! – A.E].
End of COMBINE INTERLEAVE
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
pitchfile – a binary pitch data file (.frq) as produced by REPITCH GETPITCH, REPITCH GENERATE, REPITCH COMBINE Mode 2 or BRKTOPI
formantfile – a binary formant data file (.for) as produced by FORMANTS GET
outfile – analysis file (.ana) which can be resynthesized with PVOC
MAKE combines a binary pitch data file (.frq) with a formant (spectral envelope) file (.for) to make a spectral file (.ana): .frq + .for = .ana.
Both input files must have been derived from the same PVOC settings.
COMBINE MAKE provides the standard route back to sound, following the processes of the REPITCH suite. REPITCH GETPITCH extracts time-varying pitch data from a spectral analysis file and the REPITCH functions offer a number of ways to manipulate this information. Pitch-breakpoint files can also be edited manually and converted to binary pitch files (.frq) via the function BRKTOPI, or the latter can be synthesised by REPITCH GENERATE. To get back to sound, the new pitch data file must be combined with a formant file (.for).
COMBINE MAKE is also an interesting way to combine the pitch data and the timbral (formant) data from any two arbitrarily chosen sounds.
End of COMBINE MAKE
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
pitchfile – a binary pitch data file (.frq) as produced by REPITCH GETPITCH, REPITCH GENERATE or REPITCH COMBINE, Mode 2
formantfile – a binary formant data file (.for) as produced by FORMANTS GET
envfile – a binary envelope file (.evl) produced / modified in the Time Domain or derived from an analysis file (.ana) by REPITCH ANALENV
outfile – an analysis file .ana that can be resynthesised with PVOC
.frq + .for + .evl = .ana
This function can be used to combine the pitch contour, formant envelope, and loudness envelope of different sounds, thereby transferring the characteristics of one sound onto another. MAKE2 extends the functionality of MAKE by enabling (spectral) envelope data to be used as well.The source sound could also be recreated from the data in the pitch, formant and envelope files extracted from its own spectrum. This provides a way to check on the accuracy of the extraction – small details, especially of noise sounds, often get changed.
New sounds can be engineered with this procedure by combining three different features of up to three different sounds. The pitch data file contains an extracted pitch contour, the formant data file contains the strongest, most resonant, frequency bands of the spectrum, and the envelope file contains the contour of the spectral envelope, the (changing) amplitude pattern made by the frequencies. Given that you are alert to the distinctive characteristics of your sounds, this function provides a tremendously powerful tool with which to construct new sonic material.
This is important not just from a purely sonic point of view. It is also an important tool for creating structural relationships between sounds, that is, the duplication of identical or similar shapes within a composition. This can create obvious relationships or it can be used to create subtle, hardly perceptible connections. Overall, the result is to achieve a sense of unity within the composition, even when diverse materials are being used. We are familiar with this in note-based music, where motifs and themes can have similar components, contours etc. COMBINE MAKE and COMBINE MAKE2 therefore provide a vital compositional tools in the spectral realm.
Recreating the original sound from its own pitch, formant and envelope data provides a diagnostic tool for assessing how the analysis is coping with the frequency contents of a sound.
End of COMBINE MAKE2
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
infile, infile2 [infile3 ...]– arbitrary number of input analysis files
outfile– output analysis file
COMBINE MAX studies the data in all the input analysis files. For each channel it retains only the single loudest spectral component among all the input files. In spite of the loss of the other components, the sound often remains the same to a remarkable degree. But how several sounds will merge in this way will be a little unpredictable: because data from the files which DO NOT have the loudest spectral component will simply be omitted.
The spectral characteristics of the input sounds are just merged, producing an ultra-seamless mix. The input sounds remain recognisable, just like an ordinary mix. Note that this is different from the programs which transfer the spectrum or spectral envelope of a sound. With the latter, the audible characteristics of one of the sounds are noticeably changed.
Rob Waring's wonderful hornwhivoc sound (included on the Future Music cover CD for December 1993), which we often use for demos, was made with this function. It merges a horn, a whistle and a voice – timbrally very different sounds.
End of COMBINE MAX
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
1 mean channel amp of 2 files : mean of two pitches
2 mean channel amp of 2 files : mean of two frequencies
3 channel amp from file 1 : mean of two pitches
4 channel amp from file 1 : mean of two frequencies
5 channel amp from file 2 : mean of two pitches
6 channel amp from file 2 : mean of two frequencies
7 max channel amp of 2 files : mean of two pitches
8 max channel amp of 2 files : mean of two frequencies
infile, infile2 – input analysis files made with PVOC
outfile – output analysis file
-llofrq – low frequency limit of channels to look at
-hhifrq – high frequency limit of channels to look at
-cchans– number of significant channels to compare. Default: ALL within range.
-z zeroes channels OUTSIDE the frequency range specified
This program takes 2 input sounds and calculates the average of the 2 signals, spectral window by spectral window. The average in loudness between, for example, two speech sounds, should reach some vague mean between the formants of the speech components being used. The frequency average is altogether more strange. It may, for example, convert harmonic into inharmonic spectra. Seriously wierd.
It is not really possible to predict the results of this way to combine sounds, as they are so dependent on the nature of the sources. It is best to try it with the type of source file you favour and try to build up some experimental data about what happens.
End of COMBINE MEAN
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
Example command line to interpolate pitched partials between
analysis files:
specross partials p3c.ana horn2.ana p3xh2.ana
4 5 80 5 50 2000 0.2 1 0.2
inanalfile1 – first input analysis file towards which the partials are interpolated
inanalfile2 – second input analysis file from which the partials are derived
outanalfile – output analysis file with the partials from the second file interpolated towards those of the first
tuning – the range in semitones within which the harmonics are 'in tune'; higher values are more 'forgiving' and use more of the second file (Range: 0.0 to 6.0, Default: 1)
minwin – the minimum number of adjacent windows that must be pitched, for a pitch-value to be registered (Default: 2)
signois – the signal-to-noise ration, in decibels (Default 80dB). Windows that are greater than signoisdB below the maximum level in the sound are assumed to be noise, and any detected pitch is assumed to be spurious.
harmcnt – The number of the 8 loudest peaks in the spectrum which much be harmonics in order to confirm that the sound is pitched (Default 5)
lo – the lowest acceptable frequency for a pitch (Default: 10Hz)
hi – the highest acceptable frequency for a valid pitch (Default: Nyquist/8, i.e., 2756.25 at a sample rate of 44100: sample rate divided by 2 and then divided by 8: 44100 ÷ 2 = 22050 ÷ 8 = 2756.25. The program adjusts hi according to the sample rate of the input file.)
thresh – the minimum acceptable level of any partial found, if it is to be used in reconstructed spectrum (level relative to loudest partial). A lower value is more 'forgiving' and uses more of the second file. A higher value accepts only the loudest partials.
level – the level of the output (Default: 1.0). Use if reapplying inanalfile1 to several inanalfile2s, whose relative level is important.
interp – the interpolation between inanalfile2 and inanalfile1, which can vary through time. Breakpoint time values will be scaled to the duration if inanalfile1. A lower value reduces the effect of inanalfile2, i.e., it 'lets through' more of a recognisable second file, rather than just its partials. (Default? Range?)
-a – retain the loudness contour of inanalfile2, under the contour of inanalfile1. What does this mean?
-p – extend the first stable pitch of inanalfile1 to the start of the outanalfile
The importance of SPECROSS PARTIALS is that it is focusing on pitched sounds. An interpolation is being between one sound and another: in this case, of the harmonic partials, i.e., those that are integer multiples. These are heard as pitches because the partials fuse into a single (pitched) entity. This is a welcome addition to the CDP transition programs, because the others deal with any partial, whether harmonic or inharmonic.
The harmonically-related partials taken from the pitched source in effect form a colouration, not unlike a formant. This colouration is taken from the second sound and imposed on the first, thus re-colouring it. If the thresh and interp parameters are set to high values, the partials-colouration is so strong that there is virtually nothing left audible from the rest of sound-2, only the colouration comes across to sound-1. If the thresh and interp parameters are on the low side, some of the rest of the sound will come across as well. The example command line uses lower values.
Other parameters such as tuning provide ways to fine-tune just how the harmonic qualities of sound2 are picked up. A lower value here is more 'forgiving' and is likely to use more partials.
This program provides a tool to work with recolouring pitched sounds. The various parameters enable you to define how intensely this is done, how focused the result is on the harmonically related partials, and how much else of the second sound is used, making its presence more audible in the output sound.
Although designed for use with two pitched sounds, I tried recolouring the vocal file count.ana with the sound of a horn (horn2.ana, which was horn.ana + horn.ana joined with ANALJOIN JOIN). The result was a highly modulated vocal sound with a significant amount of horn colouration – a rather interesting result.
Also see a number of other programs that deal with data transfers between sounds:
It may be useful to compare the results obtained with SPECROSS PARTIALS with those of HILITE TRACE, which retains the N loudest partials – any partial, without constraining them to harmonically related (i.e., pitched) partials.
- MORPH MORPH – to make gradual transitions between the spectrum of one file and that of another.
- NEWMORPH/2 – to make gradual transitions between the spectral peaks of one file and that of another.
- MORPH BRIDGE – to make a gradual transition from one time-specified analysis window to another.
- MORPH GLIDE – to make a gradual transition from a single analysis window to another.
- FORMANTS VOCODE – to impose the formants from the second sound onto the first.
- COMBINE CROSS – to replace the channel amplitudes of the first file with those of the second.
End of SPECROSS PARTIALS
Return to list of SPEC functions at
top of this file
Return to Main Index for
the CDP System.
Return
to Spectral Index
Example command line to impose channel amplitudes:
specsphinx specsphinx 1 in1.ana in2.ana out.ana -b0.5 -f0.5
1 Impose channel amplitudes of inanalfile2 onto the channel frequencies of inanalfile1
2 Multiply the spectra
inanalfile1 – input analysis file 1
inanalfile2 – input analysis file 2
outanalfile – output analysis file
-aampbalance – proportion of inanalfile1's amplitudes that are retained. Default = 0
-ffrqbalance – proportion of inanalfile2's frequencies injected into the output spectrum. Default = 0
-bbias – When non-zero, it adds a proportion of the original files to the output:
< 1: add inanalfile1 to outanalfile in the proportion determined by the formula-bias/(1 + bias)
> 1: add inanalfile2 to outanalfile in the proportion determined by the formulabias/(1 - bias)
-ggain – overall gain applied to the outputampbalance, frqbalance, bias and gain may vary over time
SPECSPHINX is one of three experimental programs designed to create spectra that are intermediate between two given spectra. (The other two are NEWMORPH and SPECTWIN.
Mode 1 is related to COMBINE CROSS, which replaces the channel amplitudes with those of Infile2. (Here, ampbalance sets the proportion of amplitudes to be retained.)
...
End of SPECSPHINX
Return to list of SPEC functions at
top of this file
Return to Main Index for
the CDP System.
Return to Spectral Index
Example command line to create combined spectra:
spectwin spectwin 1 in1.ana in2.ana out.ana
1 Formant envelope of inanalfile1 with the formant envelope of inanalfile2
2 Formant envelope of inanalfile1 with the total envelope of inanalfile2
3 Total envelope of inanalfile1 with the formant envelope of inanalfile2
4 Total envelope of inanalfile1 with the total envelope of inanalfile2The formant envelope traces out the envelope formed by the peaks in the spectrum.
The total envelope traces out the envelope formed by EVERY channel in the spectrum.
inanalfile1 – input analysis file 1
inanalfile2 – input analysis file 2
outanalfile – output analysis file
-ffrqint – dominance of the spectral frequencies of inanalfile2. (Range: 0 to 1. Default: 1.0)
-eenvint – dominance of the spectral envelope of inanalfile2. (Range: 0 to 1. Default: 1.0)
Note that the next three optional parameters are linked:
-ddupl – duplicate the sound of inanalfile1 dupl times, at higher pitches (Range: 0 to 8)
-sstep – the transposition step in (fractions of) semitones at each duplication (Range: 0 to 48)
-rdec – the amplitude level multiplier from one transposition to the next when creating the transposed duplications (Range: 0 to 1)
SPECTWIN creates hybrid sounds by combining the formant envelopes and/or total spectral envelopes of two files. frqint and envint specify the dominance of inanalfile2's spectral frequencies and spectral envelope, respectively. Various hybrids can be formed by reducing the effect of one of these. Roughly speaking, if envint is at maximum and frqint is at minimum, the result is a transformation of inanalfile1 (heavily influenced by inanalfile2); if the reverse, the effect is a transformation of inanalfile2.You can optionally duplicate Infile 1 dupl times at higher pitches; if chosen, the interval is set as step semitones and dec determines how much each duplicate should be attenuated (1 = no attenuation).
This function can create a wide variety of hybrids between two sounds. If both sounds are the same, the duplication feature on its own is virtually the spectral equivalent of MODIFY STACK: in Modes 3 or 4, set frqint and envint to 0, and dec to a high value. This method can be used to add extra brightness to the timbre.If one source file is longer than the other, the outfile length is that of the shorter file. (R.F.)
End of SPECTWIN
Return to List of COMBINE functions for interacting spectra 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
-ccrossover crossover is the amount of 2nd spectrum which is added to the 1st (Range 0 to 1)
crossover may vary over time
COMBINE SUM examines the amplitude levels in each channel of each of the two input analysis files. Where they differ, it adds this difference to the 1st file, as weighted by crossover. In other words, the result is infile plus the difference between it and infile2. When resynthesising, watch out for clipping.
This is a way to adjust the timbre of a sound by adding to it some part of of the timbral features which belong to another sound. Very often, the change is both major and relatively unpredictable.
Also see: COMBINE DIFF.
End of COMBINE SUM
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index
As mentioned in the text for MAKE2, there are both sonic and compositional reasons for combining the features of different sounds.
Another application is for morphing, to create intermediate stages when morphing from one sound to another. This presumes that a morph is meant to be perceptually different from a transition. The latter implies a smooth changeover from the first to the second. A morph, however, can imply a gradual and perceptible reshaping of the first with the features of the second. All of the COMBINE functions can be explored with this in mind: sums (SUM), differences (DIFF), means (MEAN) of sounds, unpredictable mixes of the loudest frequencies (MAX), combinations of pitch traces, resonant frequencies and spectral amplitude contours MAKE and MAKE2), and even an intermediate stage that interleaves the two files (INTERLEAVE).
End of Technical Discussion
Return to List of COMBINE functions for interacting
spectra at top of this file
Return
to Main Index for the CDP System.
Return
to Spectral Index