|
A Little Background: Vocoding
101
I'd wager that all electronic
musicians know what a vocoder sounds like, but no
doubt many are hazy about the technical basis of
this venerable and inscrutable effect. In a nutshell,
the vocoder imposes the time-varying spectrum of
one signal (called the "program") onto
the spectrum (either time-varying or static) of
a second signal (called the "carrier").
In its simplest and most common application the
vocoder imposes a human voice onto an instrument
sound, allowing the instrument to "sing."
History
Historically, the vocoder (short
for "voice coder") was invented in 1928
by Bell Telephone Laboratory research physicist
Homer Dudley. The original purpose of the device
was to reduce the bandwidth required for voice communications
- speech could be analyzed into a small number of
slowly varying signals at the transmission end and
resynthesized into intelligible speech at the receiving
end. While it never caught on commercially, a digital
version of Dudley's vocoder did serve as a secure
voice channel during World War II for communications
between President Franklin Roosevelt and Prime Minister
Winston Churchill. After the war the device remained
classified and was unheard by most civilians until
the 1970s, when analog channel vocoders began to
be produced by Sennheiser, EMS, Bode, Synton, and
others.
These commercial units are all
based on a two-part analog circuit consisting of
an analysis section and a syntheses section. The
spectrum of the program signal (typically a voice)
is "sliced up" by a set of band pass filters,
each allowing just a certain range of frequencies
to pass through. The signal from each of these filter
channels is then sent to an envelope follower. At
the same time the carrier signal (typically pulse
wave oscillator or a polyphonic keyboard) is also
sliced up by a second filter set, and its channels
are sent through a bank of voltage controlled amplifiers
(VCA's). The control signals from the envelope followers
are used to control the levels of the VCA's, so
that the corresponding amplitude functions of the
program signal are used to shape the amplitudes
of the carrier signal (I like to visualize it as
a jello mold for sound). Analog channel vocoders
typically have 11 to 22 bands and produce sounds
that have a "robotic" quality. Back in
1978 the top quality units, such as the Sennheiser,
cost more than most musicians made in a year.
The analog units typically offered
a choice of using an external carrier signal or
an internal oscillator and (sometimes) noise source.
When keyboard vocoders became popular in the early
1980s, manufacturers usually replaced the external
carrier input with a built-in polyphonic synthesizer
and offered an integrated microphone. While these
instruments were handy for doing singing robot effects,
they tended portray the vocoder as a one trick pony
and interest died out until the electronica renaissance.
However, in the computer music and speech research
centers the vocoder was still a matter of much interest.
Going Digital
When vocoder research moved
into the digital realm the number of channels was
no longer limited by analog circuitry. You could
have as many channels as your computer (or your
patience while waiting for the audio rendering)
could support. A variety of analysis/synthesis algorithms
have been explored over the years, but a particularly
productive method is the Fast Fourier Transform
(FFT) and its complement, the Inverse Fast Fourier
Transform (IFFT). This mathematical process allows
a high resolution analysis of a spectrum into its
individual frequency components, so rather than
the dozen or two bands of an analog vocoder a digital "phase
vocoder" can have hundreds! Newer software
vocoders like Native Instrument's Vokator are
capable of providing up to 1024 bands. Not only
does the much greater number of bands support a
higher-resolution, more natural sound quality but
by being in the digital domain it permits a wide
range of resynthesis effects such as pitch shifting,
time scaling, and formant manipulation. This level
of number crunching used to limit phase vocoding
to non-realtime functions, but more powerful programs
such as Vokator happily
take advantage of modern CPUs and run in real time
with moderate latency.
Check out Richards review of
NI's Vokator here.
Or write to sales@audioMIDI.com for
more information on programs of this type.
| 
