Cross-synthesis is a technique for creating hybrid sounds, in which the characteristics of one sound are imposed on another. CDP has three programs: VOCODE, CROSS and PUT FORMANTS. In each case, two input files are needed and the characteristics of the second (the 'modulator') are implanted on the first (the 'carrier'). Both files must have been analysed with the same PVOC settings. VOCODE VOCODE is a frequency-domain version of the classic vocoder (see NOTES below), imposing the spectral envelope of the second sound on top of the first. The number of bands is determined by how the spectral envelope is extracted: either by frequency ― how many channels are in each band; or pitch ― how many bands are set per octave. If there is a lot of high-frequency energy, the frequency analysis may be better. The rhythmic articulation of the carrier is preserved, even though the imposed spectral envelope has its own frequency-related dynamic characteristics. For example, if the carrier (Infile 1) has silence-separated repetitions of a sound, these silences are preserved (you can't impose something on nothing), but the sound when present is articulated by the imposed spectral envelope (Infile 2). MODES 1. Formants by frequency: extract the spectral envelope by frequency, using the channels parameter. The smallest available bandwidth is one analysis channel; for a 1024-point FFT this is c.43.07Hz., giving a 512-band vocoder. The bands are of equal frequency width, hence higher frequency bands have progressively smaller pitch. 2. Formants by pitch: extract the spectral envelope by pitch, using the bands parameter. The smallest available bandwidth is one semitone. The bands are of equal musical interval, hence higher frequency bands will be progressively wider in frequency. PARAMETERS channels Mode 1: no. of frequency channels used to extract spectral contour: determines width and number of formant-bands. Minimum 1
channel
per band = 512 bands for a 1024-point FFT (the default).
Higher number
gives broader bands - more general.
Range: 1-no. of channels/2
bands Mode 2:
no. of pitch
bands per octave used to extract spectral contour (greater
more
accurate).Range:
1-12
[low freq]
Optional high-pass filter: filter out all frequencies
lower than
this. Range: 5-22050 Hz.
[high freq] Optional low-pass filter: filter
out all frequencies higher than this ()Range: 5-22050 Hz.
[gain]
Optional
amplitude adjustment to the signalRange: 0.001-1-10 (CDP:
0.000002
to
10.0);
default 1 (normally < 1.0 if
required).
CROSS CROSS directly replaces the spectral envelope of the first sound with that of the second. The only control is the degree of replacement. The rhythmic articulation of the carrier is NOT preserved. To take the example given for VOCODE where the carrier (Infile 1) has silence-separated repeitions of a sound: assuming the modulator (Infile 2) is a continuous though articulated sound (e.g. bubbles), the result is also a continuous sound, matching the latter's spectral envelope. PARAMETERS [replace] degree of replacement Range: 0-1, default: 1
PUT FORMANTS PUT FORMANTS imposes or replaces the spectral envelope of the first sound, using a previously extracted spectral envelope formant file (.for). Soundshaper can extract this automatically: just select the cell containing the desired soundfile or analysis file. MODES 1. Replace formants: Spectral envelope of Infile2 (.for) replaces that of Infile1 (.ana) 2. Impose formants: Spectral envelope of Infile2 (.for) imposed on top of that of Infile1 (.ana) PARAMETERS (optional params. as in VOCODE) [low freq] Optional high-pass filter: filter out all frequencies lower than this. Range: 5-22050 Hz.
[high freq] Optional low-pass filter: filter
out all frequencies higher than this ()Range: 5-22050 Hz.
[gain]
Optional
amplitude adjustment to the signalRange: 0.001-1-10 (CDP:
0.000002
to
10.0);
default 1 (normally < 1.0 if
required).
NOTES The classic analogue vocoder is a dual filterbank of about 16 frequency bands each. The sound to be transformed, the carrier (e.g. piano), is passed through one bank while the controlling sound, the modulator (e.g. voice), is passed through the other. In each case the sound is split up into the various frequency bands. In the modulator's bank each band is passed through an envelope follower, which extracts the level in the band and imposes it onto the equivalent band in the carrier's bank. The carrier bands are then re-mixed. The result is that the frequency-related dynamic characteristics of the modulator (but not its direct signal) are imposed on to the carrier signal. The effect works best when the carrier's signal is relatively static (unrhythmic) and the modulator's is highly articulated. Note the difference between imposing the spectral envelope (VOCODE) and directly replacing it (CROSS), described under each process above. VOCODE preserves the rhythmic articulation of the carrier though that of the modulating signal is imposed on top of it. CROSS loses the articulation of the original, the time-varying spectral envelope of the modulating signal directly replacing it. PUT can implement both types, but requires an extracted spectral envelope (.for file).
|