WaveWarp 2.0 Component
      
Spectral Transformers:
Functional Description
Determines the frequency and
magnitude of the
three dominant spectral peaks in the audio input (over a selected
duration),
and sends the information out via the
control outputs. The
spectrum is computed via the FFT, and the
peaks are "tracked" in time by comparing successive
spectral snapshots. This enables the
actual location of the peaks to be determined,
irrespective of the width of the
FFT bins, thereby circumventing the
limit on frequency resolution imposed by the
finite length of the
input data buffer. See also the
Spectral 3-Peak Detector
for the simpler implementation (which is subject to
the resolution limitations
of the FFT).
Algorithm
Each successive input buffer is converted
to the frequency domain using the windowed-FFT with double overlapping.
A simple search for the maximum magnitude value
across all the
FFT bins reveals the approximate location of the dominant peak (within the
resolution of the FFT). Two further simple searches across
all FFT bins on either side of the dominant peak reveal the
approximate locations of the next two
peaks. By comparing the approximate peak locations
across successive FFT frames, the actual locations can be effectively pin-pointed,
according to the "tracking phase vocoder" technique presented in
[Moorer1]
and
[Moore] p. 570-573.
The behaviour of the tracking peak detection depends on the
settings of the FFT, as summarised in the following table.
| Parameter | Purpose |
|---|
| "Buffer length" slider |
Adjusts the length of the input data buffer,
which also defines the "Latency" (overall delay) of the
process.
The FFT buffer size is computed from the "Buffer length"
rounded up to the nearest power of 2
(for efficient FFT computation). If "double padding" is selected, the FFT buffer size is
doubled (after the rounding) to improve the smoothness of the spectrum between successive
FFT bins (but without increasing the underlying frequency resolution).
The input data buffer
is windowed (using a selected profile), then
extended to the length of the FFT buffer by
padding with zeros (on either side). Successive input data buffers are overlapped by a factor
of 2.
The appropriate choice of "Buffer length" is a trade-off between
frequency resolution (improved with longer buffers) and temporal resolution
(degraded with longer buffers).
For a given application, the most suitable
choice usually depends on the characteristics of the audio signal.
The buffer length should be long enough that the dominant
peaks are located within separate FFT bins. Thereafter, the peak-tracking technique
will narrow down their location inside the bins, so the buffer length does not need to
be further increased.
|
| "Input gain" slider |
Adjusts the overall amplitude of the input signal,
before computing the FFT.
|
| "Bin masking" slider |
Determines the "masking" zone (expressed in terms
of the number of FFT bins) around each detected peak. The search for the next dominant
peak will exclude all FFT bins within the masking zone. This reduces the
possibility of false detection caused by "spectral smearing"
in the vicinity of a peak (arising from the fact that the peak may not
lie within a single bin, for a given FFT resolution).
|
| "Window type" selection |
Selects the profile of the windowing function applied to the input data.
|
For an introduction to the Discrete Fourier Transform and the FFT,
see, for example,
[St] sections 4.1 and 4.2.
For further introductory information (with emphasis on audio applications), and
for discussions on spectral measurements, zero-padding,
windowing,
and the Short Time Fourier Transform (STFT) for audio applications,
see
[Roa] p. 1084-1112
and
[Moore] p. 61-111.
Signal Implementations
| Audio signals | Control signals | Description |
|---|
| Single input mono | 6 outputs |
The first and second control outputs contain the frequency (in Hz) and the magnitude,
respectively,
of the dominant spectral peak of the mono input.
The third and fourth control outputs contain the frequency (in Hz) and the magnitude,
respectively, of the second (or third) dominant peak.
The fifth and sixth control outputs contain the frequency (in Hz) and the magnitude,
respectively, of the third (or second) dominant peak. Note that the
second and third dominant peaks are not necessarily ordered in terms of
the control ouput sequence (this reduces the computational burden).
|
| Single input stereo | 6 outputs |
The first and second control outputs contain the frequency (in Hz) and the magnitude,
respectively,
of the dominant spectral peak of the average of the stereo input channels.
The third and fourth control outputs contain the frequency (in Hz) and the magnitude,
respectively, of the second (or third) dominant peak.
The fifth and sixth control outputs contain the frequency (in Hz) and the magnitude,
respectively, of the third (or second) dominant peak. Note that the
second and third dominant peaks are not necessarily ordered in terms of
the control ouput sequence (this reduces the computational burden).
|
Related components:
Example DrawingBoards illustrating usage:
|
