![]() |
|
![]() |
FREQUENCY DOMAIN
Many CDP processes take place in the Frequency Domain - that is, they process the frequencies that make up the sound.
When we hear a pitched note, we hear one pitch with a tone quality (timbre). The tone's complex waveform can be analysed as a spectrum of component frequencies, or partials, that make up the sound:
In this analysis of an organ note, the most prominent spectral peaks show the first 12 harmonics of the sound: the peaks are equally spaced and the partials are simple whole-number multiples of the lowest, or fundamental, frequency, which is why we we hear the sound as having a definite pitch.
The spectrum is shown as a plot of frequency (Hertz, abbrev. Hz) against dynamic level (deciBels, abbrev. dB). Above about 4KHz, the peaks are indistinct, or randomly spaced: this indicates the presence of noise. In bell-like sounds, the partials are said to be inharmonic, as they have no clear fundamental and may also be unevenly spaced in frequency:
![]()
The dynamic contour of the spectrum is known as the spectral envelope. In the above examples, this would includes all the peaks and troughs, but might also be interpreted more generally as a decay in loudness from low frequency to high.
In practice, most sounds do not have a static spectral envelope, but a dynamic one: the relative strengths of the partials are changing all the time. (The images above are averaged over a given time.) In acoustic sounds, it is especially common for the weaker higher partials to decay first.
An alternative term for spectral peak is formant. Within CDP, formant files (.for) are in fact an extraction of the time-varying spectral envelope. In phonetics, the term 'formant' denotes a particular type of resonance: a fixed frequency region in which any partials are amplified. Speech formants are most noticeable in vowels: in speaking or singing a vowel we are creating two (or three) formants which define the quality of "ee" or "oo", irrespective of which pitch we sing. (For further discussion, see the article: What is a formant?)
PHASE VOCODER (PVOC)
CDP uses the Phase Vocoder (PVOC) a version of the Fast Fourier Transform (FFT) to convert sound files into frequency analysis files (suffix .ana) (also known as spectral files). PVOC analyses the spectrum by splitting the spectrum into equal frequency bands, called channels:
![]()
It also makes successive frequency analyses in time, using overlapping time-slices, or windows, of a few milliseconds:
There is a trade-off between frequency and time more channels give better frequency resolution but worse time resolution. A typical analysis file has 1024 points, giving 513 channels (1024/2 + 1 for 0Hz). Each channel has two points: one for frequency, one for amplitude.
The diagrams above depict a 1024-point analysis in which the audio sample-rate of 44100Hz/1024 gives a channel bandwidth of c. 43.07 Hz. The analysis sample-rate here is 44100Hz/128 = 344.53125 Hz. Put another way, a new spectral analysis starts every 2.9 msecs (1000 msecs/344.53125 = 2.9024944) and each window's time-frame of 8 steps or 1024 samples is 23.2199552 msecs.
You can read more about the Phase Vocoder in the CDP documentation:
T. Wishart: The CARL Phase Vocoder (PDF version: here )
R. Dobson: The Operation of the Phase Vocoder (PDF version: here )SPECTRAL MENU
In Soundshaper, spectral processes are found in the Spectral Menu (or in the menu Processes | Spectral):
For an overview of all the spectral processes, see CDP Spectral Processes.
AUTO-CONVERSION
Natively, CDP spectral processes are in mono and you would normally process one analysis file to create another one. However in Soundshaper, .wav to .ana conversions are done automatically behind the scenes, using the current PVOC settings, which you can alter. (Default PVOC values are sufficient for most purposes.) You simply choose a spectral process.
STEREO and MULTI-CHANNEL PROCESSING
Soundshaper splits a stereo or multichannel soundfile into its component channels and processes these independently. For example, the output of a stereo sound put through a spectral process might be shown as ~A_1_c1.ana, representing channel 1. The channel indicator in the Play Panel displays the number of channels.
You can also have a different parameter set for each channel.
PITCH and FORMANTS
CDP also has processes which use pitch files (.frq) and formant files (.for), both of which are derived from frequency analysis files (see especially PITCH-PROCESSES). Again, Soundshaper uses automatic conversion wherever possible.
RETURN to top of page |