Digital-Formant Synthesizer for Speech-Synthesis Studies
Digital-Formant Synthesizer for Speech-Synthesis Studies
Digital-Formant Synthesizer for Speech-Synthesis Studies
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
9.4, 9.8 Received 1 November 1967<br />
Diital-<strong>Formant</strong> <strong>Synthesizer</strong> <strong>for</strong> <strong>Speech</strong>-<strong>Synthesis</strong> <strong>Studies</strong><br />
LAWRENCE R. RABINER<br />
Bell Telephone Laboratories, Incorporated, 3fltray tlill, A;ezc, Jersey L7974<br />
Recent work on speech synthesis by rule has led to the design and evaluation of a new digital <strong>for</strong>mant<br />
synthesizer. The synthesizer was simulated on a digital computer at a 20-kHz sampling frequency. The<br />
improvements over previous synthesizers include: an excitation sonrce <strong>for</strong> the unvoiced compmeut of<br />
voiced fricatives; a method <strong>for</strong> producing a voice bar <strong>for</strong> the closure period of voiced stops; and a simple<br />
desigu <strong>for</strong> the higher pole correction network that exploits the properties of sampled data systems. Spectrographic<br />
examples of synthetic speech are shmvn in order to illustrate these improvementS. A brief discussion<br />
of some advantages and disadvantages of serial and parallel synthesizers is presented.<br />
INTRODUCTION<br />
N recent lears, an extensive amount of time and ability of a terminal analog synthesizer to produce high-<br />
ef<strong>for</strong>t hats .gone into designing and evaluating qnalily natural-sounding speech has been demonstral ed<br />
machines capable of converting tow-in<strong>for</strong>mation-rate repealedly, u,-ø thus motivating ef<strong>for</strong>ts to increase the<br />
control signals into speech. -4 hnprovements in channel capability of this machine and to make it more versatile.<br />
and <strong>for</strong>mant vocoders have been in part responsible <strong>for</strong> Its potential as a speech-research tool is high.<br />
this resnrgence of interest in speech synthesizers. :,-7 This paper is concerned with at digital simnlation of<br />
Furthermore, tire general availability of both large and a serial terminal-analog synthesizer. The synthesizer<br />
small digital computers 8. and advances in digital and was sinrelated on a large computer (GE 45) usin<br />
analog hardware have contributed in large meastu'e to FORTRAN I¾. The general structm'e of the synthesizer<br />
the widespread interest in speech synthesizers. in relation to the production of various classe<br />
Among the more common speech synthesizers is the sounds is examined closely, and improvements over<br />
terminal-analog synthesizer. Besides serving as the previous synthesizers are discussed in detail. These imsynthesizer<br />
<strong>for</strong> a <strong>for</strong>mant vocoder, a terminal-analog provements include a voiced fricative excitation netsynthesizer<br />
is often used to generate material <strong>for</strong> experi work, a new design <strong>for</strong> the higher pole-correction network,<br />
md a simple method of producing a voice bar <strong>for</strong><br />
R. S. Tomlinson, "SPASS An Improved Terminal-Analog voiced stop consonants. A discnssion of some ad-<br />
<strong>Speech</strong> <strong>Synthesizer</strong>," J. Acoust. Soc. Am. 38, 940(A) (1965).<br />
vantages and disadvantages of parallel and serial<br />
G. Fant, J. Martony, U. Rengman, and A. Risherg, "OVE II<br />
<strong>Synthesis</strong> Strategy," Paper I:5, <strong>Speech</strong> Conminn. Seminar, realizations of a terminal-analog synthesizer conclude<br />
Stockhohn (1962).<br />
this paper.<br />
a j. L. Flanagan, "Note ou the Design nf Terminal-Analog<br />
<strong>Speech</strong> <strong>Synthesizer</strong>s," J. Acoust. Soc. Am. 29, 306--310 (1957).<br />
* C. H. Coker and P. Cummiskey, "On-Line Computer Control<br />
of a <strong>Formant</strong> S.xmthesizcr," J. Acoust. Soc. Am. 38, 940(A)<br />
(1965).<br />
s C. H. Coker, "ReabTime <strong>Formant</strong> Vocoder, Using a Filter<br />
Bank, a General-Purpose <strong>Digital</strong> Computer and an Analog <strong>Synthesizer</strong>,"<br />
J. Acoust, Soc. Am. 38, 940(A) (1965).<br />
B. Gold, "Techniques <strong>for</strong> <strong>Speech</strong> Bandwidth Compression,<br />
Using Combinations of Channel Voc(>ders and <strong>Formant</strong> Vocoders,"<br />
J. Acoust. Soc. Am. 38, 2 10 (1965).<br />
M. R. Schroeder, "Vocoders: Analysis and <strong>Synthesis</strong> of<br />
<strong>Speech</strong>." Proc. IEEE 54, 720-734 (1966).<br />
8 p. Denes, "On Line Computing in <strong>Speech</strong> Research," Proc.<br />
Intern. Congr. Acoust. 5th Liege (1965), Paper A23.<br />
822 Volume 43 Number 4 1968<br />
ments in speech synthesis and perception.0.. The<br />
I. GENERAL DESCRIPTION<br />
of speech<br />
A termired-analog synthesizer models the speech producing<br />
mechanism, including the vocal tract, excitation<br />
0 K. Xakata, "<strong>Synthesis</strong> of Nasal Consonants by a Terminal<br />
Analog <strong>Synthesizer</strong>," J. Radio Res. Labs. 6. 243 254 (1959).<br />
o L. Rabiner, "<strong>Speech</strong> Svnthesis hv Rtde l An Acoustic Domain<br />
Approach," Bell System 'lrech. J. 47,' 17-37 (1968).<br />
u j. N. Holmes "Notes on <strong>Synthesis</strong> \V0rk," <strong>Speech</strong> Transmss<br />
on Lab. Qnart. Progr. Rept.', Stockholm (April 1961).<br />
2 W. J. Strong, "Machine-Aided <strong>Formant</strong> Determination <strong>for</strong><br />
<strong>Speech</strong> Ssmthesis," J. Acoust. Soc. Am. 41, 1434 1442 (1967j.
DIGI'FAL-FORMANT NYNTHE.IZER<br />
f T t-<br />
I"m. 1. Blck (liegram .f sl)cech synthesizer.<br />
sources ,red r,tdiation impedance. The fieorv of repre- tontrol sigmd is set to t,ttc the pitch impul_-e generator<br />
senLing vocal-tract lelmvior in tel-mS O[ ils Irtnsmis- Otltpnl to the shaping nehvork.<br />
sion properlies, I) a trausfcr fnnclion has been
L. R. RABIXER<br />
F o COMPOiEI T Figure 2 shows the relevant details of this network.<br />
OF VOICED<br />
(Certain components of the synthesizer of Fig. 1 have<br />
FILTER NET .ORK<br />
been omitted from Fig. 2 <strong>for</strong> simplicity.)<br />
The unvoiced excitation is produced as follows. The<br />
pitch pulses excite a resonator tuned to the first <strong>for</strong>erant<br />
of the voiced cmnponent of the fricative. This resonator<br />
Bt RESONATOR<br />
is the first-order pproximation to the transfer function<br />
of voltame velocity (the signal of interest in Fig. 2) from<br />
the glottis through the point of constriction of the vocal<br />
tract. A threshold level (Vn) is subtracted from the<br />
output of the resomUor, and the restlit is half-wave<br />
rectified. These operations model the physical fact that<br />
turbulence is not produced until the volume velocity of<br />
the airflow exceeds a threshold value.<br />
'['he output of the half-wave rectifier modulates the<br />
GENRATCR NOISE J b'VOlCEO outpnt of the noise generator, producing a pitch-<br />
COMPONEN1<br />
OF OlCES synchronous excitation <strong>for</strong> the unvoiced component of<br />
the fricative. The final unvoiced component is produced<br />
l'lc.. 2. Excitation netxmrk <strong>for</strong> voiced fricatives. by feeding this excitation into the fricative network<br />
(i.e., the lower branch of the synthesizer). The voiced<br />
generator, a zain control, a multiplier, a frication pole component is prodneed by exciting the <strong>for</strong>e,ant network<br />
and zero network, and a shaping network. The gain by the pitch pulses.<br />
control AFRIC provides means <strong>for</strong> controlling the over- Figure 3 shows spectrograms of synthetic and natural<br />
all unvoiced output level.<br />
versions of the voiced fricatives/z/and/5/- The syn-<br />
The input to the multiplier from the voiced-fricative thetic speech was produced eutirelv bv rule. m The<br />
excitation network is constant except during a voiced natural speech is presented <strong>for</strong> comparison purposes. A<br />
fricative. For a voiced fricative, the input to lhe multi- careful examination of these spectrograms shows the<br />
plier is a signal that pitch s3nchronously modulates effects of the pitch-synchronous modulation during the<br />
the unvoiced input. The details of this technique are fricatives fro' bolh the synthetic and the natural speech.<br />
explained in the next Section.<br />
The presence of the low-frequency voiced components<br />
The fricative pole and zero network is simply a cas- <strong>for</strong> both the synthetic and natural versions can also be<br />
cade of a complex-conjugate pole-pair resonator and a<br />
seen in Fig. 3. The major difference between the comcomplex-conjugate<br />
zero-pair resonator. The pole and<br />
zero positions are variable (FPOL, FZER). The shaping<br />
panion spectrograms is that the lower <strong>for</strong>mants of the<br />
network i, used to provide high- and low-frequency natnral speech are weaker than the lower <strong>for</strong>mants of<br />
emphasis, a, well a,to acconnt <strong>for</strong> radiation characteristics.<br />
A cascade of a differentiaLor and a low-frequency<br />
resonator is used here.<br />
The middle branch of the synthesizer is used to produce<br />
a voicebar <strong>for</strong> the closure interval of voiced stop<br />
consonants. This branch consists simply of a gain<br />
control (AVB) that adjusts the level of the pitch pulses<br />
6- {<br />
produced at the output of the ,aptrig network in the<br />
upper branch.<br />
The output ,eech is produced by adding the output<br />
from each of the three branches.<br />
II. VOICED FRICATIVE EXCITATION NETWORK _ I<br />
The voiced-fricative excitation network that connects s-<br />
the output of the pnlse-shaping filter to the lower<br />
branch of the sx nthesizer is used to model the produc- .z a-<br />
I,,<br />
tion of the unvoiced component of voiced fricatives) :-'<br />
HUMAN ,%øEEC H<br />
O. Fujimura (personal communication, 1067) has indicated<br />
o<br />
that G. Rosen used a modulation scheme to produce voiced frlca- o-<br />
tires on D \VO [G. Rosen, "Dynamic Analog <strong>Speech</strong> Synthe- aS u<br />
sizer." J. Acouq. Soc. Am. 30, 201-210 (1958)]; however, no<br />
x titten record of his techniques appears to be available.<br />
I'h(. 3. Spectrographic examples of voiced fricatives.<br />
824 Volume 43 Number 4 1968<br />
P, HZ 3-<br />
2-<br />
HUMAN SPEECH<br />
! I<br />
SYNTHETIC & Z ae -- SYNTHETIC
DIGITAL FORMANT SYNTH ESIZEI(<br />
the synthetic versions, indicating a need <strong>for</strong> low-fi-equency<br />
de-emphasis.<br />
The intelligibility of the voSced fricatives was measured<br />
in a <strong>for</strong>mal test, ils part of a larger-scale evaluation<br />
of synthetic consonants./" The relevant resttits are that<br />
/z/ and /5/ were identified carteeth' 100% of the time<br />
in prestressed and poststressed position in VCV tests.<br />
There were 15 possible responses <strong>for</strong> each stimnlus in<br />
these tests. The voiced fricatives /v/ and .'8./ are not<br />
included in this discussion because the unvoiced component<br />
was not found necessary <strong>for</strong> their synthesis.<br />
Hence, they were synthesized in the same way ils vocalic<br />
sounds.<br />
No <strong>for</strong>mal tests were conducted to determine the<br />
quality of the synthetic voiced fricativcs, bul in<strong>for</strong>mal<br />
comments by listeners were highly favorable.<br />
III. HIGHER-POLE CORRECTION NETWORK<br />
I"m. 4. Pole positions <strong>for</strong> a five-pole 10-kHz synthesizer.<br />
+24<br />
- t 2 ST RESPO 'NSE 0 ' AC<br />
0<br />
I 2 TUBE 5 POL ANA<br />
36<br />
48<br />
O I 2 3 4<br />
FREQUENCY ( k H Z ) --'<br />
l"JO. 3. <strong>Digital</strong> and aliatog approximations to transfer function<br />
of an acoustic tube, open at me end and closed at the other end.<br />
sampling fi'equency i.e., the pole at 500 Hz represents<br />
The higher-pole correction network is used to com- analog poles at 500, 10 500, 20 500 Hz, etc. Hence, the<br />
pensate <strong>for</strong> the effects, Oil the low-frequency- spectrum, infinity of higher-order poles is guaranteed by using this<br />
of the missing higher-order poles in the usual approxi- method. The density of the higher-order poles is also<br />
mation to the vocal-tract transfer function. In an correct, thus automatically ensm'ing the necessary<br />
analog synthesizer, a network approximating the trans- higher-pole correction. The efi'ectiveness of the digitill<br />
fer function of the higher-order poles is used. In a digital higher-pole correction is shown in Fig. 5. Here we see<br />
synthesizer, however, the design of the higher-pole car five-pole digital and five-pole anah)g approximations to<br />
rection is relativeh' simple. Additional resonators, or the transfer function of a straight acoustic tube model<strong>for</strong>mant<br />
networks, placed at appropriately chosen fre- ing the vowel/'a/'. No tnalog higher-pole correction is<br />
quencies, provide the necessary higher-pole correction. used other than the additional fixed <strong>for</strong>mants at 3500<br />
To illnstnte this method, consider it synthesizer and 4500 Hz. '['he bandwidths of the digital ired analog<br />
with a 10-kHz sampliug frequency. Assume the vowel poles are increasing with frequency- tmodeling the ob-<br />
/a/is to be synthesized and the pole positions are 500, served data <strong>for</strong> speech), whereas the bandwidths of the<br />
1500, 2500, 3500, 4500, ... (2n-+-l) $00 Hz. As shown poles of the tube transfer function are constant. The<br />
in Fig. 4, we place the three resonators of the <strong>for</strong>mant high-frequency behavior of the analog approximation is<br />
network at 500, 1500, and 2500 Hz. The higher-pole significantly different from the tube trinsfer function,<br />
correction network consists of resonators at 3500 and indicating the need <strong>for</strong> additionM higher-pole correction.<br />
4500 Hz. Owing to the periodicity, in frequency, of The advantages of the automatic higher-pole correction<br />
sampled data systems, the pole at any' frequency repre of the digital approximation are apparent.<br />
sents an infinity of poles with it spacing e(lual to the The modifications of the higher-pole correction network<br />
<strong>for</strong> a 20-kHz sampling frequency are straight<strong>for</strong>ward.<br />
Instead of inserting fixed resonators at just<br />
z - PLANE<br />
35(10 and 4500 Hz, additional resonators at 5500, 6500,<br />
7500, 8500, and 9500 Hz are necessary. There will be 10<br />
poles in the upper half-plane of Fig. 4, instead of five as<br />
<strong>for</strong> it 10-kHz syuthesizer; however, the poles are at the<br />
same frequencies as be<strong>for</strong>e. For a 20-kHz synthesizer,<br />
TAB1,E I. Resonator frequencies and bandwidths.<br />
Resonator C[: (ttz) Q BW (Hz)<br />
1:4 35011 2(I 175<br />
I":, 4500 16 281<br />
b' 5500 12 458<br />
I"7 6500 9 722<br />
l"a 7500 6 1250<br />
Fo 8500 4 2125<br />
P'o 9500 2 4750<br />
The Journal of the Acoustical Society of America 825
kHz<br />
kHz<br />
4--<br />
2--<br />
5--<br />
4--<br />
3--<br />
2--<br />
1--<br />
O-- !<br />
El U T<br />
T<br />
1<br />
I<br />
L. R. RABINER<br />
t t'1 I<br />
, I<br />
A N S I S T O R R A D IO<br />
S E L DOM GET<br />
I'm. 6.1..SI)ectrograms of synthetic speech.<br />
R I C H<br />
the b,ndwidfi-, of the additional poles are.reater than citin the upper branch of the sx nthesizer !)v the pitch<br />
the bandwidth of the similar poles <strong>for</strong> a 10-kHz sxn- impulses<br />
thesizer. The bandwidths used are seen iu Table L The Voiceless stop consonants are produced using the<br />
values chosen were determined from data by l)unn. s upper and h)wer branches of the synthesizer. The time<br />
interval following the period of closure can be broken<br />
IV. SYNTHESIS CHARACTERISTICS<br />
SPEECH SOUNDS<br />
OF THE down into three senents. During the firsl seg-ment,<br />
there is a short burst of friczttion uoise. This uoise is produced<br />
using the lower branch of the synthesizer. The<br />
In order to produce the various clases of speech<br />
output of the fricalion enerator is multiplied by m exsounds,<br />
only certain portions of the synthesizer need<br />
ternal ,unplitude control (AI"RIC) and then multiplied<br />
generally be used at one time. Vowels, semivowels, by a fixed constant. The output of the second multiliquids,<br />
and glide are produced using only the upper plier is fed into the fricative pole and zero network, and<br />
branch o[ the x nthesizer. The pitch-impulse generator<br />
then into the shaping network.<br />
is the source of excitation <strong>for</strong> these sounds.<br />
The second segment of a voiceless stop is the aspira-<br />
Voiced stop consonants are produced using both the<br />
tion ptmsc. Aspiration i. produced bv gating the output<br />
upper and middle branches o[ the synthesizer. The voice of the frication generator to the upper branch of the<br />
bar is produced bx gating the output of the pitch- synthesizer. The third segment of the voiceless stop,<br />
impulse generator to the shaping network ,red lhen the voiced transitional phase, is px oduced in exactly the<br />
multiplying the output of thi.,, network by an extern,'fi same m,mner as <strong>for</strong> voiced stops.<br />
amplitude control (AVB). This provides a low-fre-<br />
The voiceless fricatives are produced in the same way<br />
quency 1ow-enerzy wave<strong>for</strong>m whose characteristics are as Ihe [rication noise <strong>for</strong> voiceless stop consonanls. As<br />
similar to voice bars observed in real speech. The<br />
shown above, a voiced fricative is synthesized as two<br />
tnmsitional sebment of voiced stops is produced by exseparate<br />
components. The voiced cmponent is produced<br />
in a manner similar to the voiced components of<br />
olher speech sounds--i.e., by excitin the upper branch<br />
]I. K. Dunn. "Methods of Xleauring Vowel Formalit Bandwidths,"<br />
J. -cout. ,qoc. Am. 33. 1737 1746 (1961).<br />
826 Volume 43 Number 4 1968
DIG[TAL-FORMANT SYNT[fES[ZER<br />
of the synthesizer by pitch impulses. The unvoiced component<br />
is produced by modulating the frication generator<br />
output by a pitch-synchronous wave<strong>for</strong>m--the<br />
output of the voiced-fricative excitation network. The<br />
modulated wave<strong>for</strong>m is then passed through the lower<br />
branch of the synthesizer and added to the voiced component<br />
to produce the voiced fricatlve. An/h/ sound<br />
and whispered speech are produced in a manner similar<br />
to that <strong>for</strong> aspiration.<br />
A. Spectrographic Examples<br />
In order to illustrate the capabilities of the synthesizer,<br />
we show, in Fig. 6, spectrograms of two synthetic<br />
utterances. The control signals used to drive the synthesizer<br />
were determined by rule using an auxiliary progrant.<br />
There<strong>for</strong>e, there are no original speech samples to<br />
present <strong>for</strong> comparison purposes.<br />
I I<br />
These spectrograms exhibit the various classes of<br />
speech sounds. The upper spectrogram shows both<br />
voiced sounds and unvoiced sounds. Of most interest in<br />
Fio. 7. Parallel terminal-analog synthesizer.<br />
this spectrogram are the phoneraes in the word lransis- ticated parallel synthesizer would account <strong>for</strong> <strong>for</strong>mant<br />
lor. I)uring the/t/, both the frication noise burst and phaseø-0).<br />
aspiration are visible. The effects of the nasal pole and The serial synthesizer presents several advantages<br />
zero <strong>for</strong>/n/' are noticeable in the region of 2 kHz, where over the synthesizer of Fig. 7. Individual amplitudes of<br />
there is little energy present. Finally, a comparison is each of the resonances do not have to be determined <strong>for</strong><br />
available between the voiced fricative /z/ and its voice-<br />
, /<br />
a serial synthesizer. For research on speech synthesis by<br />
less counterpart /s/. The effects of the pitch-synchronous<br />
modulation are easily identified by the vertical<br />
rule, this is a significant advantage, as it reduces the<br />
striations through the noise of/z/. Furthermore, the complexity of the rules. The vowel spectra from the<br />
lack of a low-fi'equency voiced spectrum <strong>for</strong>/s/is easily parallel synthesizer of Fig. 7 contain extraneous zeros,<br />
observed.<br />
whereas a serial synthesizer produces spectra containing<br />
The lower spectrogram exhibits two interesting features.<br />
The voice bars of /d/ in "seldom" and /g/ in<br />
only poles. These zeros generally fall at frequencies<br />
between the resonances and may be perceptible, and<br />
"get" appear as low-frequency vertical striations during hence a corrupting factor. Because of the presence of<br />
their respective stop gaps. The voice bars appear as zeros in the spectra, parallel synthesizers generally have<br />
light lines indicating that they are of low energy. no higher-pole correction networks, relying instead on<br />
Finally, the affricate /ti/ of "rich" appears as a stop the zeros to provide the low-frequency emphasis. This<br />
gap followed by a burst and ending in a period of frica- is an nnreliable technique because: the positions of the<br />
tion energy. The frication energy is similar to the energy zeros move as the resonances are varied.<br />
of the /'/ phoneme. An affricate is synthesized as a<br />
The parallel synthesizer has two advantages over a<br />
stoplike sound followed by a fricativelike sound on this<br />
serial synthesizer. Noise generated in the parallel synsynthesizer.<br />
thesizer propagates additively rather than multiplicatively,<br />
as in a serial synthesizer. For simulations where<br />
B. Parallel versus Serial Realization<br />
small register lengths are required, this advantage is<br />
particularly valuable. For a given signal-to-noise ratio,<br />
The discussion in this paper has been concerned pri- signal sizes in a parallel synthesizer are smaller than in a<br />
marily with a serial tenuinal-analog synthesizer. A serial synthesizer. The second advantage is the ability<br />
parallel synthesizer is generally realized by gating either<br />
of a parallel synthesizer to reproduce consonant spectra<br />
of two sources (a noise or pitch-impulse generator) to a<br />
accurately through independent control of <strong>for</strong>mant<br />
series of parallel branches, each containing a resonator<br />
amplitudes. Hence, <strong>for</strong> synthesis from spectrograms or<br />
and a gain control, and adding the output of each<br />
in a <strong>for</strong>mant vocoder, this degree of freedom can be of<br />
branch to produce the speech. Figure 7 shows a simple<br />
realization of a parallel synthesizer l,ø (a more sophisgreat<br />
value.<br />
J-. Holmes, I. Mattingly, and J. Shearme, "<strong>Speech</strong> <strong>Synthesis</strong><br />
by Rule," Language and <strong>Speech</strong> 7, 127-143 (1964).<br />
PITCH<br />
GENERATOR [<br />
EECH<br />
J. L. Flanagan, "Recent <strong>Studies</strong> in <strong>Speech</strong> Research at Bell<br />
Telephone Laboratories (II)," Proc Intern. Congr. Acoust., 5th<br />
IJage (1965), Paper A22.<br />
The Journal of the Acoustical Society of America 827
L. R. RABINER<br />
The choice of a parallel or serial synthesizer is determined<br />
primarily by its application. For research in<br />
speech synthesis by rule, the serial synthesizer has been<br />
very useful.<br />
have been included to demonstrate the capabilities of<br />
the synthesizer.<br />
ACKNOWLEDGMENTS<br />
V. SUMMARY<br />
I am indebted to Dr. Cecil Coker of ]3ell Telephone<br />
Laboratories and Professor Osamu Fujimura of the<br />
A new design <strong>for</strong> a digital terminal-analog synthesizer Research Institute of Logopedics and Phoniatrics <strong>for</strong><br />
has been presented and discussed. The new features many valuable discussions on the design of a speech<br />
include a voiced fricative excitation network, a method synthesizer; to Dr. _[ames Flanagan, Professor Kenneth<br />
of producing a voice bar, and a simple higher-pole cor- Stevens, and Professor William Henke <strong>for</strong> sugõestions<br />
rection network based on the properties of sampled data and ideas about speech synthesis; and to Dr. Harry<br />
systems. Spectrographic examples of synthetic speech Levitt <strong>for</strong> valuable criticisms of the manuscript.<br />
828 Volume 43 Number 4 1968