The resonant circuits of the synthesizer can be arranged either in
parallel, the outputs of the various circuits being summed to produce
the final output, or in series, the output of each resonant circuit
being fed into the next. The series synthesizer is a closer
approximation to the acoustical behavior of the vocal tract, and
incorporates in its design Fant's (1956) observation that if certain
assumptions are made about the glottal source spectrum and the formant
bandwidths, the relative amplitudes of vowel formants can be predicted
from their frequencies. Thus a series synthesizer requires fewer
parameters and can be expected to produce more natural vowels. On the
other hand, parallel synthesizers are far more flexible, and simplify
synthesis strategy for sounds with complex spectra, like voiced
fricatives. The relative merits of parallel and series synthesizers
are far best summed up by Flanagan (1957). Lawrence's (1953) PAT was
the first example of a parallel resonance synthesizer; Fant's (1958)
OVE II, the first full-scale experimental series synthesizer. Highly
reliable resonance synthesizers of both types are now
available. 3
A parametric control scheme should have made synthesis by rule simpler,
since the parameters to be specified are precisely the dimensions of
speech in terms of which it is convenient to state acoustic rules.
Ingemann (1960), in fact, reformulated her rules for use with the
Edinburgh series version of PAT (Anthony and Lawrence 1962). But in
order to control a resonance synthesizer, some means of changing the
parameter values dynamically is required. At first this was
accomplished with a function generator: for example, parameter
functions for the Edinburgh PAT were represented in conductive ink
on parallel tracks of a moving plastic belt (Fourcin 1960). But
applying the rules (as distinct from stating them)
was if anything
more troublesome with a function generator than with the Pattern
Playback. Fortunately, digital computers now began to become available
for phonetic research. Kelly and Gerstman (1961) demonstrated that
the computer not only could apply a set of rules (i.e., calculate
the parameter values) quickly and accurately but also could be used
to simulate the synthesizer itself. 4 Other
investigators showed that
if an actual, rather than a simulated synthesizer is used, the
computer could also play the role of function
generator. 5
__________
3. Other parallel resonance synthesizers are described in Borst
1956, Holmes et al. 1964, Mattingly 1968b and Glace 1968; other
series synthesizers are described in Coker 1965, Tomlinson 1965,
Liljencrants 1968, Kacprowski and Mikiel 1968, Kato et al. 1968,
Dixon and Maxey 1970, Shoup, pers. comm.
4. The advantage of a simulation is that it can be completely
reliable and accurate, and the design of the synthesizer can be
readily modified; the disadvantage is that an extremely powerful
computer is required and such computers are too expensive to permit
extended real-time operation. Recent simulations of resonance
synthesizers (all series) include those described in Flanagan
et al. 1962, Rao and Thosar 1967, Rabiner 1968, Saito and
Hashimoto 1968.
5. On-line transmission of stored parameter values can be performed
by a laboratory computer at a cost low enough to permit the
investigator to experiment at length; it is easy to program other
convenient facilities such as routines for editing or displaying
the stored parameter values. Schemes of this sort include those of
Tomlinson (1965), Denes (1965), Coker and Cummiskey (1965), Scott
et al. (1966), Mattingly (1968b). Off-line control schemes,
in which the computer produces a record, such as a paper tape,
which is then used to control a function generator, are also
practical, though less convenient (Holmes et al. 1964; Iles 1969).
|