CreamWare Noah Vocodizer Manual page 4

How does a Vocoder actually work?
A typical vocoder contains two filter
banks: an analysis bank and a synthesis
bank. As the name implies, the analysis
bank takes the incoming voice signals and
analyzes them to extract frequency and
timing information, creating a "signature"
for each filter in the bank. These signatures
are later applied to an identically tuned
filter bank in the synthesis section, and
used to process a secnd, synthesized,
signal (or, it could be an entirely unrelated
external input signal). In both filter
sections, the filters are divided into an
equal number of frequency bands.
After the analysis phase, the vocoder
examines each individual analysis filter
output with an envelope follower to extract
dynamic level information. The envelope
follower then generates control signals
corresponding to the changing volume
levels.
The component signals from the synthesis
filter bank (whose filters are tuned to the
same "signatures" as the analysis filters)
are multiplied by the control signals from
the envelope follower. In other words, the
Vocodizer
dynamic level progression of the signals
filtered by the synthesis section match
exactly the dynamics of the filtered signals
of the original voice input. Finally, the
synthesized signal components are mixed
to produce the final output. The result is
vocoder output that produces a
synthesized sound with the character and
articulation of the original input voice. The
vocoder is often used to produce the
"talking synthesizer" effect made popular
by artists such as Stevie Wonder.
To guarantee the highest quality results,
the Vocodizer implements Voiced/
Unvoiced Detection. During the analysis
phase, the signal is examined for its tonal
vs. noise content. Vowels, such as "A"
are identified as tonal (voiced), and
consonants such as "S" as noise
(unvoiced). Depending on the results of
this analysis, either the synthesizer signal
or a noise signal is passed on to the
synthesis filter bank. A noise signal
(noise, in this case, referring to a signal
containing all frequencies in appropriate
proportions) can help to reproduce the
sibilant portions of the original signal.
Often the synthesized results produced
by a traditional vocoder do not contain
enough high frequencies. The
Vocodizer's "noise substitution" strategy
is able to produce more convincing
results.
If you want, you can use the sibilant
portions of the original signal for the
unvoiced source instead of a random
noise signal. In this case, the vocoder
section filters the input signal in such a
way that only the higher frequency
components remain, letting you use the
original "S" sounds to replace noise as
the unvoiced source. In most cases,
though, the broadband noise signal is a
more appropriate source for unvoiced
signal components, as the output
already has a synthetic quality to it, and
the original signal, if used, tends to stand
out too much. But this option can come
in handy sometimes, and in the long run
it's all a matter of taste and
appropriateness for the situation.
Contents Index
4