the speaking machine of wolfgang von kempelen (homer dudley and t. h. tarnoczy)

SPEAKING MACHINE OF WOLFGANG VON KEMPELEN 151

noise calculations are given in column 12, and values of Zr from Eq. (41) or

Zr = B,+AB,+ 8+ K- •7o' (41)

are given in the last column.

APPENDIX 2: HEARING LOSS FOP. SPEECH

In this-appendix the relationship between the hearing loss for speech •4 and the hearing loss audiogram will be considered. Let t•i be the hearing loss at the frequency f for a pure tone. It is the ordinate in the audiogram. If we consider t•i has the same effect upon the threshold level as an attenuation -R from the

flat response system, then by analogy to Eq. (23) the hearing loss for speech/t, is given by

10-•,n0= fo•øG110-•llOdf. (99) • H. Fletcher, "A method of calculating hearing loss for speech

from an audiogram," J. Acous. Soc. Am. 22, 1 (1950).

If we consider only the octave frequencies 125, 250, 500, 1000, 2000, 4000, and gOO0, then the following equation is approximately correct.

•,= - 10 log{EWe10 -•/'ø} (lO0)

where k takes the successive values of 125, 250, 500, 1000, 2000, 4000, and 8000. The weights are given by

W• = Jo.7• G•df. For 125 c.p.s. W=0.000, 250 c.p.s. W=0.003, 500 c•p.s. W=0.104, 1000 c.p.s. W=0.388, 2000 c.p.s.

. W=0.395 , 4000 c.p.s. W=0.106, 8000 c.p.s. W=0.004. So for most purposes one needs to consider only the

'four frequencies 500, 1000, 2000, and 4000 and use weights 0,1, 0.4, 0.4 and 0.1.

For a fairly flat audiogram it is approximately correct to take an average of the hearing loss at 500, 1000, and 2000 c.p.s.

THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA VOLUME 22, NUMBER 2 MARCH, 1950

The Speaking Machine of Wolfgang von Kempelen* Homer DrrDL•

Bdl Talephone Laboratories, Murray Hi!l, New Jersey

Alqn

T. H. T^•ocz¾

Biophysi•al Laboratory of the Institute for Anthropology of the Museum of Natural History, Budapest, Hungary (Received October 17, 1949)

The physiological motions involved in speaking can be indicated to the eye or to the ear. For the eye suitably chosen symbols may be written to indicate the physiological positions assumed in forming each sound; for the ear synthetic sounds may be produced by motions in a mechanism built to simulate the speech organs. The degree of phonetic success may be estimated in the case of the visible symbols by listening to sounds formed when the indicated physiological processes are carried out, and in the case of the speech- simulating mechanism by comparing the synthetic speech produced to normally spoken speech. Signifi•nt advances along both the visual and the aural lines are described from earliest times down to the present.

Wolfgang yon Kempelen produced the first speaking machine worthy of the name around 1780. This paper gives his background, a description of the apparatus he built, and a discussion of the methods used in producing the various sounds, fitting his work into the over-all picture of speech-imitating devices from the speaking of idols of ancient times down to the automatic electrical reconstructing of speech in the vocoder. For portraying to the eye the physiological characteristics of speech there are discussed the more outstanding methods from claimed symbolic alphabets of ancient languages down to the recent spectrographic visible speech.

OWARD the end of the 18th Century a Hun- garian, Wolfgang Ritter von Kempelen, or, in

Hungarian, Kempden Farkas Lovag, first built a complete and, on the whole, a surprisingly successful speaking machine. Speech was formed by manipulation

* Orally presented before the Acoustical Society of America, May 5, 1949, New York, by Dudley with original draft by Tarnoczy. The paper here, in general, follows the oral presenta- tion including a set of figures and also other material not in the original draft.

of mechanical elements simulating the essential parts of the human vocal system. In 1791 he published a 456- page book, • illustrated with 25 plates, describing his observations on human speech production and his experiments during the two decades he had been working on his speaking machine. The appearance of his book was a great social event. Introductory to the

• Mechanismus der menschlichen Sprache nebst der Beschreibung seiner sprechenden Maschine. Also published in French at the same time (1791).

Downloaded 22 Sep 2011 to 130.91.140.232. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp

152 HO3,IER DUDLEY' AND T. tl. T\RNOCZV

book are listed, as subscribers, 122 names of outstanding people of Austria, Russia, Italy, and Hungary.

•,on Kempelen was born on January 23, 1734, in P( vsony, a city in western HungaD', now Bratislava, Czechoslovakia, and died in Vienna on March 26, 1804. [ : reached high government position in the Habsburg mt,narchy becoming, in 1767, Aulic Counselor of the (_hamer of the Domain of Empress and Queen Maria '1 her es• He traveled widely over Europe. He made the

ins 6f the fountains in Sch•inbrunn and later the

le,igns of the Royal Castle of Buda. He organized the •, ool manufacture in South-Hungary. He was a skilfful engineer and a genius in organization.

'•-•e question a ises as to what motivated van Kempelen to attempt the building of any speaking machine, for no one had, so far as known, attempted this difficult task in the centuries preceding the day m 1769 he started a job that, off and on, would take over two decades to complete. The answer in part, no doubt, is that he had some interest in the problem of speech by deaf-mutes. But, more fundamentally, there were several factors favorable to the development of a speaking machine by a man of his ability and inclination at that particular time. Pervading all times has been the basic importance and significance of speech. But, in additioIl to this general urge, there was stirring abroad in van Kempelen's day an aroused scientific curiosity in the wake of the Reformation, and from this there had arisen gradually the faint beginnings of physiological phonetics as a science. These three factors will be discussed briefly before returning to his book.

Speech is of such basic importance that civilization as known today could not exist without it, yet like the air we breathe, it so envelops us that we take it for granted. The continuing importance of speech in its

Fla. 1. Automaton chessplayer of van Kempelen. Courtesy of Oxford University Press.

application to modem civilization is indicated by the industries built upon speech since yon Kempelen's day. Man's unaided voice carries less than a mile at a speffd of only 12 miles per minute, a speed actually somewhat less than that of a fast jet plane of today. But the telephone industries carry the spoken word thousands of miles and at speeds of millions of miles per minute; radio industries provide for audiences numbering millions of people at a time; phonograph industries preserve speech for unborn generations to hear. Trans- mission of the written word has been accomplished by the telegraph and more recently the facsimile industries. •nd even in yon Kempelen's times there were industries handling the printed word which have expanded enormously in the intervening years to flood us with books, magazines, newspapers, maps, folders, pam- phlets, advertisements, etc. Religion, education, the professions, in fact all organized society, depended then as now upon the spoken and written word. By speech man raised himself to a position above and distinct from the lower animals. Accordingly, in ancient times, man took speech as a symbol of his divine origin and assumed his gods were, somehow, speaking gods. Naturally then, the priests tried to make their idols appear to speak directly to the people. For this purpose speech was piped in from a concealed priest to make words issue forth from the mouth of the Oracle of

Orpheus on the Isle of Lesbos? In one revelation, this oracle accurately predicted the violent death of Cyrus the Great in his expedition against the Scythians. In the Middle Ages, a thousand years later, Roger Bacon and others built small talking heads of bronze and wood as models of ingenuity with concealed tubes bringing in a speaker's voice but without intent of superstitious deception.

Von Kempelen was born in a time of aroused scientific curiosity. Galileo had passed on less than a century earlier. There was stirring a healthful skepticism which demanded that truth be sought not in a blind faith but by experimental methods of cut-and-try. This scientific vigor manifested itself strongly in the design of auto- mata to produce motions of various sorts when energized as by. winding a spring. Thus a Frenchman named Vaucanson • built a man-like figure, or android as it was called, that played a flute with all the complicated motions needed for the lips and fingers. I-Ie later made an automaton for simultaneously playing a shepherd's pipe held in one hand while beating a tamhour with the other hand. He also developed a miraculous duck, perhaps in some ways the most extraordinary automaton ever constructed. It moved its wings and walked in a natural manner. It drank water, muddling in the

'act. It would take corn from one's hand and swallow it

with a complete simulation of the digestion process

: The ScieMifi• Papers of Sir Cltarles Wlteatslone (1879). Sir David Brex•ster, Letters on A•atural }[aglc (1832).

aSci. Am. 24, 32 (1871). An Account of the •[echanism of an Automaton, etc. (translation by J. T. Desaguliers) (1742). Downloaded 22 Sep 2011 to 130.91.140.232. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp


aided by chemical means. Another Frenchman, Le Droz, made up a writing child while his son made a bullfinch that would jump up from a snuff box, wag its tail, spread its wings, pour forth a melodious song, and then dart down into the box as the lid closed.

Now yon Kempelen was also a skilled mechanician. In the spirit of mechanical ingenuity of the times he built* in 1769 his famous "chess automaton" (Fig. 1) with a turbaned Turk playing an alm6st unbeatable game of chess seated at a desk on which there was mounted a chess board. The device was exhibited over

Europe in the next four years by yon Kempelen and after his death by Maelzel in Europe and later in America until burned in a fire in 1836. Before the exhibition the cloak at the back of the Turk was lifted

and, several at a time, various compartments of the desk were opened up to show their apparent emptiness. When everything was closed, with a lot of noise from inside machinery, the Turk laid down a long-stemmed pipe and started a chess move. Von Kempden himself remarked that this chess player was not a true automaton, its only. outstanding feature being the skill of the deception. The deception apparently consisted in having a skilled player conceal himself in the cabinet and play the game from information received in the upward displacement of a small iron ball under each square on which there was placed one of the chessmen, each of which contained a strong magnet. The chess player is said to have beaten Napoleon in one of its games. In 1821, when the player was being exhibited in London, Robert Willis, noted for his later researches on synthetic vowel production 5 wrote a 40-page booklet 5 on the chess player. His attempted analysis does not agree entirely with Murray's explanation of the automaton mechanism.

The difficulty of the problem, the mechanical ingenuity, the breadth of view and to some extent the feel and daring for showmanship manifested by von Kempden in the chess player are characteristics shown in his development of the speaking machine started in the same year, 1769. But the chess automaton was completed in six months while the speaking machine occupied von Kempelen for much of the time for over twenty years before he published his results in 1791.

This aroused and growing spirit of experimental in- vestigation in science in general naturally led to ques- tions as to the physiology of speech production, thus laying the foundation of experimental phonetics as a science. A century before von Kempelen's studies, Baron Franciscus Mercurius ab Helmont published a book 7 in Latin contending that the Hebrew alphabet

' H. J. R. Murray, A History of Chess (Oxford University Press, London, 1913), pp. 876--7.

s Robert Willis, Trans. Camb. Phil. Soc. 3, 231 (1829). ½ Robert Willis, An Attempt to Analyze the Automaton Chess

Player of M. de Kempden with an Easy Method of Imitating the Movements of that Calebrated Figure, with 10 Plates (1821).

*Baron Franciscus Mercurius ab Helmont, Alphabeti vere Naturalis Hebrai•i Brevissima Ddineatio (1667).

Fro. 2. Hebrew letter.M as a tongue position according to Helmont.

was a "natural" alphabet in that the letter symbols represented the actual tongue positions and so should be useful in teaching speech to the deaf-dumb. Von Kempelen reproduced four of Helmont's illustrations, one of which is shown here as Fig. 2. The tongue position is for M, the 13th letter of the Hebrew alphabet, pro- nounced Mere, as indicated in Hebrew below the picture of the head. The figures in the border of the headband are related forms for the letter M, those numbered "4," for instance, being copied from ancient coins. Von Kempelen criticized Helmont, pointing out that for the sound of M, the tongue position is of little importance but that the significant positions are in the open nasal passages and the closed lips, although the latter are shown open by Helmont. The small element ahead of the 'distorted tongue is part of the Hebrew letter M but seems to have no phonetic significance.

A more serious representation of alphabetic char- acters by the use of phonetic symbols portraying vocal tract positions was published in a book s a year later by Bishop John Wilkins in England. Figure 3 is a copy of page 378 of his book. Here are illustrated the pictures of the head with vocal positions for 34 sounds consisting of eight vowels and 13 pairs of voiced consonants with unvoiced counterparts. Corfiparing the consonants in the list with those given by Fletcher ø one notes that both lists contain four voiced and four unvoiced stop consonants, four voiced and four unvoiced fricatives, and five voiced semivowels. Wilkins differs in showing

• Bishop John Wilkins, An Essay Towards a Real Character and a Philosophical Language (1668).

9 I-I. Fletcher, Speech and Hearing (1929), p. 6.


154 HOMER DUDLEY AND T. H. TARNOCZ¾



a complete set of unvoiced semivowels hi, hm, hn, hng, and hr, and omitting the three transitionals y, w, and h. The unvoiced semivowels are found in some languages with Welsh containing all five. He lists y as a vowel, probably considered w as a vowel (we= fiE) and really provided a multiplicity of h sounds with his five unvoiced semivowels. He showed only eight vowels as against 11 in the Fletcher list. In the upper right-hand corner for each of the 34 sound illustrations are Wilkins'

proposed phonetic symbols. In the upper left portion of Fig. 3, nine heads are shown with no sectioning in a three-by-three arrangement for those sounds having significant lip positions. In each of the other 25 illustrations, a section is shown to reveal the position of vocal parts within the mouth and throat. Some physically significant symbols used are a circle for the rounded lips of the o sound, a pair of dosed lips for p, closed lips with a semicircle or small wave to represent vocal cord action for 13, a straight horizontal line through the mouth to represent the flow of air through the mouth for the consonants where there is a relatively free flow, and a slightly curved line at the top to represent the flow of air through the nose as in the case of m, hm, n, and hn.

Wilkins, in the same book, page 376, shows a sort of syllabary reproduced here as Fig. 4. Consonants following a vowel in a syllable are shown in small letters as listed in column 1; those preceding the vowel are shown by capital letters as listed in column 9. Column 2 gives the consonant symbols for indicating the consonants alone. The row numbered ! gives the six vowels he used and row 2, their symbols. Columns 3-8 give vowel- consonant combinations and columns 10-15, the consonant-vowel combinations. He employs a small circle placed high, middle, or low for the first three of his vowels and a semicircle similarly placed for the other three. His consonant list of Fig. 4 includes h, w, and y, though these are not in the illustrated list of Fig. 3. The reason for including them here is that they combine with the vowels. At the bottom of the figure is the Lord's Prayer written with these symbols.

To round out the picture of phonetic portrayal, we note that since yon Kempelen's time, considerable further advance has been made. In particular, Alexander Melville Bell tø worked out a set of symbols he termed "visible speech," showing in minute detail the complete vocal action in producing not only the speech sounds but also whispers, whistles, sobs, grunts, clicks, hisses, sighs, coughs, sneezes, kisses, and all other sounds producible by the human vocal mechanism. A detailed explanation of the symbols is also given by his son, Alexander Graham Bell. n Figure 5 •2 shows his basic "universal alphabet," the sounds from it used in the

•o Alexander Melville Bell, Visible Speech--The Science of [lni- versal Alphabetks (D. Van Nostrend Company, Inc.• New York• 1867).

u Alexander Graham Bell, Me•hanisra o] Speezh (1907), second edition.

t• From Explanatory Lecture on Visible Speech (1870).

English alphabet, and two diagrams to define the chief consonant and vowel symbols, respectively. In the universal alphabet table, consonants are represented in column.• 1-4; the vowels in 6-8; glides in 5; throat sounds and modifiers in 9 and 0. The first six rows of

consonants are the unvoiced ones, with the voiced counterparts given in the next six rows with short inside bars added to represent voicing. The diagram at the left shows three cord positions, bar for voicing, a circle for wide open as in producing the sound h of column 9, row a, i.e., 9a, and an X for the closed glottis in 9c. A partial opening as in whispering is shown by an oval in 9b with a bar added to show hoarse vocality as in 9h. The point of constriction (aperture) is represented in the diagram by the concave quarter of the three- quarter circles. The aperture may be at the back of the tongue, the front of the tongue, the point of the tongue or the lips producing the consbnant columns 1-4 which are called back, front, point, and lip consonants, respectively. In rows b and h of the consonant symbols there is a minor aperture in the quarter opposite to the main aperture as represented by a pair of small three- quarter circles; the indicated consonants are termed back-mixed, front-mixed, point-mixed, and lip-mixed. In rows c and i the main quarter of the symbol is indented to represent dividing the 'air stream as in the sound f (4c); such consonants are called back- divided, front-divided, etc. Mixture combined with division gives the mixed-divided consonants represented in rows d and k. Rows e and I have the three-quarter circle closed by a straight line representing a "shut" position as in the stop consonants; thus le is the back- shut consonant k, 11 is the back-shut voice consonant g, etc. In the final rows f and m of the consonants, the wavy line from the uvula in the diagram .refers to closing the nasal passages with the uvula to give the nasal consonants called back-nasal, etc. The meanings of the symbols for the tongue positions for vowels are given in the right diagram. An upright bar indicates a vowel as in columns 6-8 throughout, while the small horizontal bars in the lower six rows indicates round

vowels, i.e., vowels spoken with round lips. The back, front, and mixed positions of the tongue are represented by dots or hooks at the left, right, and left plus right of the vertical vowel bar, respectively. The hook indicates that the voice channel is opened wide, the dot not so wide. The elevation of the tongue is indicated by the dot or hook positions on the vertical bar, a low elevation by the bottom position, a high elevation by the top position, and a medium or mid-elevation by the combination of one dot or hook at the bottom and the

other at the top. Thus u as in pull listed as 6k would be ß defined as a high-back-wide-round vowel. This explanation covering 84 consonants and vowels, many not to be found in any language, is considered adequate for showing how the physiological production of the various speech sounds is indicated in great detail by Mr. Bell.

The International Phonetic Association, founded in


156 HOMER DUDLEY AND T. H. TARNOCZY

!

3

6 7

8•

FIG. 4. Wilkin's English language syllabary for a philosophical language.

•.( •1,( IVLT '(L /d LIJ, I01"/

• I,.('q /I/), •.( "l/d /l '1, I/U •q,



1886, drew up an International Phonetic Alphabet in 1888 providing symbols for presumably all speech sounds but without special physiological connotation as in the case of symbols shown in the present paper. Recently Potter •a and his co-workers have developed visible speech analysis with the sound spectrograph, portraying the sound spectrum automatically, thus replacing assumed positions of the vocal parts with objectively recorded physical characteristics of actually spoken speech. The words "visible speech" from such a spectrogram are shown in Fig. 6a, while Fig. 6b shows the set of manual alphabetic symbols used in an experimental training program to portray the distinguishing characteristics of speech sounds as revealed by spec- trograms.

In the century preceding yon Kempelen, there was much speculation by linguists, physicists, psychologists, and speech teachers on the mechanism of speech. The understanding of physical science was advancing from philosophical speculation to scientific experimentation with physical apparatus. Von Kempelen, as stated earlier, started working on his speaking machine in 1769 and continued to 1791 at least. Others were also

active in this period. Thus, in 1779 the Imperial Academy of St. Petersburg offered its annual prize for explaining physiological differences in, and making apparatus for, producing the five vowel sounds/A (father), E (they), I (machine), O (note), and lJ (crude). These are the long vowel sounds as used on the Continent and so, for consistency, will be used here. The prize was won by Christian Gottlieb Kratzenstein, TM born in Wernigerode, Germany, who became a Professor of Physiology first at Halle and later at Copenhagen. He made five tubes as shown in cross section in Fig. 7. These tubes roughly approximated the size and shape of the vocal passages when set to produce the different sounds. All were energized by free reeds except the I tube which was blown into directly in the fashion of an organ pipe. It is interesting to note that Robert Willis, s showed that the shapes of Kratzenstein's tubes were not important as the required resonances could be obtained from a single pipe the length of which was adjusted for the different sounds, and that, in fact, the vowel series could be covered alternately forward and in reverse as the pipe length was increased.

From the foregoing background we return to a review of the highlights of yon Kempelen's book. This book is

•a Potter, Green, and Kopp, Visible Speech (D. Van Nostrand Company, Inc., New York, 1947). Some other writers have described apparatus for automatic writing of the symbols by the voice but the symbols lack in definiteness and uniformity so that they can hardly be classified as alphabetic in the stage of development described; See, for example, J. B. Flower, "phonographic alphabet" in "The true nature of speech," Trans. A.I.E.E. 35, 213-48 (1916) and J. Dreyfus-Graf, "Steno-sonographic alphabet" in "Le sonographe: elements et principes," Schweizer Archiv 14, 353-62 (1948).

• "Tentamen coronaturn de voce," Acta Acad. Pettop. (1780). The complete article is in "Sur la naissance de la formation de• voyelles," J. de Physique 21, 358-80 (1782).

divided into five chapters, thus: Chapter I, pp. 1-28--Speech in General; Chapter II, pp. 28-56•Origin of Speech; Chapter III, pp. 57-177--Vocal Parts and Their Functions; Chapter IV, pp. 178-387--Sounds of European Languages; Chapter V, pp. 388-456--The Speaking Machine.

The first two chapters give brief discussions of speech and its origin in quite general fashion. In the third chapter he examines into which bodily parts contribute to the forming of speech sounds and how these parts function normally and under faulty conditions. He thus explains the voice, considering particularly the func- tioning of the lungs, trachea, glottis, nose, mouth, tongue, teeth, and lips. In the fourth chapter he lists most of the European alphabetic sounds. He then proceeds for each sound and by groups to consider what would be the simplest hand-operated mechanism to produce fair imitations of these sounds. First he observes his own

vocal system as objectively as possible to determine experimentally the relative mouth and tongue-channel openings for the vowels a, e, i, o, and u, with these results:

Vowel sound Mouth opening Tongue-channel opening a $ 3 e 4 2

i 3 1 o 2 4

u 1 5.

The numbers show relative sizes, increasing from 1 to 5. He extends his observations to fit into this tabulation

seven other vowels for a total of 12. Then he proceeds to the consonants. Figure 8, yon Kempelen's Plate XII, illustrates his method of practical visualization and adaptation as he reduces applicable vocal parts to the simplest manually operable mechanism for a group of sounds. In this figure, for making the voiced explosive sounds B and D, he substitutes for the mouth a wooden box, for the lips a pair of hinged wooden shutters, for the tongue a hinged wooden flap, operable by a string, and for the air supply a tube to which could be fitted a reed for voicing. The two top subfigures are for the B sound before and after the start of the explosive emission of air; the two lower ones, for D. These cut-and-try designs were useful to yon Kempelen in clarifying his understanding of the mechanics of sound production by the human mechanism. His final speaking machine sometimes used less idealized methods as even in this

case of the B and D sounds, as will be explained later. Von Kempelen in the fifth and final chapter of his

book describes the steps of building his speaking machine, his tests as he went along and finally how to produce the different sounds and combine them into speech. He began with a search for a suitable sound source for imitating the tone from the vocal cords. The most natural-sounding source he found after ex- amirting many musical instruments was a drone reed from a bagpipe. He first tried to produce the vowels with a bell-shaped mouth attached to such a reed as



Consonan Po$igion$ Vowel Positions

THE UNIVERSAL ALPHABET.

* The Marginal Numbers and Letters may be used, instead of the Visible Speech Letters, to express the mechanism of sounds in common type. The followlng examples show the English, Scotch, and Irish pronunciations of the words ' Visible Speech"

Fro. Sa..

shown in Fig. 9. Instead of using the sliding plate shown at the bottom he usually placed his hand on or near the bell mouth in various positions to form the different sounds but only obtained non-vowel tones and later vowel-like tones with a characteristic "ah" sound no matter how his hands were placed. The results were not very satisfactory so for his second model he built a console as shown in Fig. 10 with 13 piano-like keys con-

trolling air from a common bellows to separate passages each containing a reed, thus using individual reeds for the different voiced sounds. He tried rectangular boxes such as the two at the left and then round boxes such as the four at the right. He writes he obtained some good vowel distinctions for the first time, forming a fair a, o, and u. He also obtained fair consonant sounds for p, m, and 1. With these he formed simple words like



THE ENGLISH ALPHABET.

[The Italic letters are the English equivalents of the Visible Speech Letters in the •orr•ondln• sections of the Universal Alphabet.]

i 2 1• 4 5 6 7 • 8 9 0"•

a he a ro• eel ]• aeeen• a

b s sh wh up alell •otd• b

'd t•in near -t•n* -ces* •11 d

e k t p ask -al* alr• e

• ink hin• lamp l arm• sir an f

• ] zeal azure way' oldll

k ] t•en pull -ure • k

1 , 2 3 4 fi 6 7 8 9 0

The sounds marked * occur only in unaccented syllables; as in lr3[(•l• (mention); D•{•f•If•-places; 13{t•%• (fatal; D•I•t;•/(pleasure); Ot•%t•œ (history; JO],•.•/(orator}.

The' glide' 5 a heard between a vowel and r; as in hero, airy, fief, gloor, &c. The sounds in ' ale' and ' old' include the ' glides' 5 c. and 5/. Thus :•

C•O (ale); ]'l• (old). R final or before a consonant, as in air, arm• &c•, is the ' Point-Glide' •a[

Thus :• {It (air); l• (arm).

Accent is always on the first syllable unless othenvhe •Cl•[}(0[•O (ex'9ressed). The mark is placed 1•['13•}• (be'fore) the syllable to which it G)I'•I• (reSfers). •

FI•. 5b.

Fro. 5. The "universal alphabet" of Alexander Mdville Bell with symbol-defining diagrams and illustrations in English. (See opposite page for Fig. 5a.)

"papa" and "mama" but noticed two troubles; first, the sounds did not blend together in a natural way, and second, vowels in particular came on rather explosively, thus adding a k-like sound.

He decided that to overcome the first trouble he

should do away with his multiplicity of reeds, producing all voiced sounds from the same reed. Then, to prevent the explosive oncoming of vowels, he lined the reed and the edge it beat against with thin soft glove leather

around the closure on both sides. He experimented with an adjustable wire to change the effective length of the vibrating reed and thus alter the pitch but he found it

ß difficult to obtain a variable pitch so he satisfied himself with a monotone, stating that he left this improvement to later workers in the field. In this connection, it is of interest to note that J. R. Ewald ta developed

• J. R. Ewald, "Zur kOnstruktion yon polsterpfdfen," Pfitigers Archiv f. die gesarnte PhYsiologie 1•2, 171 (1913).



some springed cushions to vibrate in a pipe in close resemblance to the human mechanism and that the

artificial laryn_.x invented by R. R. Riesz •6 gave a fair range of pitch variation according to the air pressure applied to the vibrating reed, increasing pressure giving increasing pitch; also that the method yon Kempelen tried had long been successfully applied to pipe organ reeds.

Having thus failed in his first two attempts to make a speaking machine along the lines represented in Figs. 9 and 10, yon Kempelen then started anew. This third effort resulted in his final speaking machine, the essentials of which are shown with a scale indicating size in Fig. 11, a reproduction of his Plate XXV. The bellows X shown in part at the top is used to set up an air pressure in "wind-box" A. With by-pass keys s and sch closed, the excess pressure in box A can only be reduced by leakage as discussed later and by passage of air against the reed edge, thus setting it in vibratory motion.

A brief description will be given of how the different sounds are produced according to yon Kempelen. The operator rests his right arm on the bellows X and pumps it with an up-and-down motion, speech being produced on the down motion. The fingers of the right hand are set to operate the special consonant controls marked r, sch, n, m, and s. The left hand is placed palm inward before the opening bc of bell C. The vowels are produced by working the bellows with the right elbow while blocking the nostril-imitating tubes m and n by fingers of the right hand, with the left hand set in such position before C as listening and practice indicated best for the particular vowel being produced. For sound a the hand is kept distant from the mouth opening; for e the hand is hollowed slightly with its bottom edge against the mouth and its top edge about one inch away; for o the top of the hollowed hand should be

about one-half inch from the mouth; for u the hand is held •at with the opening of the mouth reduced to a minimum short of stopping the reed vibration; but with the opening greater than for i; for i the flat hand is placed. tightly across the mouth opening and the index finger then crooked, so that there appears at the second knuckle a small opening, more air pressure being required for this vowel than for the others. He says the positions for other vowels such as the umlauts are intermediate to the given positions and can easily be located with a small amount of practice. His description of positions infers he tried to modify a second resonant frequency of the vowel spectrum as does his considera- tion of the two openings, tongue-to-palate and mouth.

An over-all picture of the consonants made on yon Kempelen's machine is obtained by comparing the ones he claimed with a list of the 24 consonants used

in English as given by Fletcher :9 Class Produced \'ot produced

Semivowels 1, m, n, r, ng Stops p, b; t, d; k, g; ch, j (judge) ¾ricatives f, v; s, z; sh, zh; th (thin), th' (then). Transitionals h, w, y (German j)

Von Kempelen also produced the sound of ch (ich) found in German but not in English for a total of 19 consonant sounds. Of the six sounds listed as not produced, none are mentioned by yon Kempelen. Of these six sounds, only ng is found in the German language.

Of the four semivowels listed as produced by yon Kempelen sound I was made like the vowels but with the left thumb curving inside the rubber mouth to correspond to the way I is produced in normal speech with the tongue arched to divide the air stream in the mouth. The sound of m was produced by closing the mouth with the left hand while leaving open both nostrils and sound n by leaving open only one nostril of the speaking machine. To produce r he adjusted his

I V I. .• I E• LE. II $ P rr CH Fro. 6a. Recent visible speech spectrogram (Potter). Courtesy of D. Van .N'ostrand Company, Inc.

•6 R. R. Riesz, "Description and demonstration of an artificial larynx," .[. Acous. Soc. Am. 1, 273 (1930). Downloaded 22 Sep 2011 to 130.91.140.232. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp


p b k g t d p•¾ be key go to gay

f v • 5 s z • 3 t•[ d 3 for vote thin then .see zoo she azure church judge

h he ohead whe

m n r] w j r 1 me no sing we you reed let

eve it hate met at esk father not all obey foot boot

--- l • •. 3 g' el al al clu ou

about up word word say I boy out go new

Fza. 6b. Manual s3 mbols for alphabet of recent visible speech (Potter). Courtesy of D. Van Nostrand Company, Inc.

positions for the following vowel and then depressed the key marked r which pushed a wire into contact with the vibrating reed, thus giving a rattling effect

which was claimed to produce a trilled r that was not perfect but better than many people could make.

Of the six stop consonants produced, he made p by Downloaded 22 Sep 2011 to 130.91.140.232. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp

162 HOMER DUDLEY kND T. H. TARNOCZY

F[o. 7. Schematic representation of Kratzenstein's five vowel synthesizers (from Young's Vatural Philosophy, 1845).

closing all openings and then, when pressure was built up, releasing the hand suddenly from in front of the mouth. This sound was rather weak at first so he provided an auxiliary storage bellows under the left side of the wind-box, not visible in Fig. 11 but clearly shown in the next figure. He found that building up the pressure in the "wind-box" and the mouth for sound p resulted, upon release of pressure, in sufficient air passing the reed to produce a voicing effect so he placed a small by-passing brass tube as shown in Fig. 11 near the sch key between the wind-box A and the bell mouth C. This, he says led to a good p sound. Adding voicing to p gave b. For t, d; k, g, he did not provide separate means as would seem indicated but after considerable experimentation he decided that he could modify the p and b sounds in a way, that he does not describe in detail, to give tolerable resemblances to the desired sounds.

Of the five fricatives and three transitionals produced, he made f from the leakage of air with all openings closed while exerting strong bellows pressure. He made h in the same way but with the mouth left open and with less pressure on the bellows; German ch was made like h but with the bellows pressed slightly harder but not enough to vibrate the reed. He made v like f except that a small escape of air was permitted at the mouth opening bc, sufficient to vibrate the reed. With less air but a larger percentage of vibrational power from the reed, the sound became w. The sound of s was made by depressing the key marked s with everything else closed; in this case the depression of key s opened a by-pass to air at the side of the reed with the air escaping through the small funnel shown under the s key, this funnel being designed of a resonant size and shape to make a hiss like the sound s. a•dding a little voicing to s gave z. Similarly, sh (always written sch by yon Kempden) was made by depressing the key marked sch, whereupon the air was by-passed to the other side of the reed through the escape tube shown at the bottom which tube was designed for an sh sound resonance. Adding voicing produced y, i.e., German j. a,n intermediate amount of voicing could have pro-

duced zh, one would think, but yon Kempelen does not mention zh at all.

The by-pass tube from wind-box A to bell C evidently wasted some of his air for he tells how he could only produce short combinati&•s of sounds in connected speech, from a complete depression of the bellows

B

B

D

FIG. 8. Von Kempelen's schematic showing essential features for mechanizing production of B and D sounds.



although th• bellows had sLx times the air capacity of the lungs. Because bf this, he was limited to producing short phrases at a time such as "Leopoldus Secundus," while the human vocal system can produce phrases several times longer on a single expiration of the breath.

Summarizing, of the 19 consonant sounds yon Kem- pelen tells of producing, we find six (p, b; t, d; k, g)

Fra. 9. Von Kempelen's first vowel synthesizer.

Fro. 10. Von Kempelen's console for some vowels and consonants.

jr :

: \l

'z ; I '

Fro. 11. Von Kempelen's final speaking machine.

were stop consonants made with an arrangement designed for p; five, consisting of the three fricatives, f, v, and German ch, and two transitionMs, h and w, were made by using escaping air having a hissing sound similar to f; seven (s, z; sh, y; m; n; r) were made by switching in five different resonant passages; and one, 1, was made, as in human speech, by splitting the air- stream in the mouth.

This completes the description of yon Kempelen's work on his speaking machine. While he mentions his belief that the final machine can easily be fitted with keys like a piano (the second model, Fig. 10, had keys but not a single reed) and says in conclusion that if he found time to improve the machine, he would continue his writing, explaining what he had done, we have no further record of any such writing from the date of the publication of his book in 1791 to 1804 when he died. One must admire the patience of the man who worked 22 years on his speech machine before completing it to the point of wanting to describe it in writing and also the perserverance of one who twice discards some two years' work and starts over completely afresh.



He states that in three weeks time a person can make astonishing progress in playing the machine if he limits himself to the Latin, French, and Italian languages. German is much harder because of the prevalence of consonants in German.

Von Kempelen was not only a skilled mechanician but he had an acute and observant ear for speech sounds as well as a lot of common sense. Some of his

first-hand observations .reveal a true understanding of the manifold nature of speech and its interpretation. The similarity of all vowels when sustained for a considerable time is a characteristic the modern worker on

speech is likely to observe, perhaps with annoyance; its disappearance in dynamic speech is cheering. The difficulty of starting a vowel sound without an accom- panying explosive effect is also well known. He again observes the difficulty of getting a substitute for the vocal cords that sounds at all voice-like. He mentions

that selecting a high pitched voice is an advantage because a child's voice is not criticized. He made .use of the lack of sameness from voice to voice when he

modified the p-b sounds to obtain his tolerable resemblances to the t-d and k-g sounds. Finally he ob- served that people would interpret sounds much more easily when given some clue beforehand.

In conclusion, we shall round out the picture by

briefly mentioning some significant further develop- ments of speech synthesis from his time on. About this same time, Kratzenstein and also Abb• Mical of France are said to have built speaking machines which they exhibited in Paris with pin-cylinder drives as in music boxes. These were presumably inferior to von Kem- pelen's in quality of speech produced. Professor Wheat- stone, from von Kempelen's description, built a speaking machine •? which he demonstrated in the Dublin meeting of the British Association for the Advancement of Sciences in August, 1835. This is shown schematically in Fig. 12. In 1846 a certain Professor Joseph Faber of Vienna demonstrated a speech roaching s advertised as "Euphonia" in the Egyptian Hall, Piccadilly, London. The device is pictured in Fig. 13. A ticket costing one shilling permitted the bearer to hear the performance consisting of ordinary speech, whispered speech, conversation, and the singing of airs ending with "God Save the Queen" (Victoria's reign). This machine is said. to have been a big improvement over Yon Kempelen's, particularly from the standpoint of having a variable pitch that permitted singing.

Helmholtz •ø with a series of tuning forks, Koenig •ø with a shaped siren, Miller • and Stumpf • with sets of pipes, Preece and Stroh •a with geared wheels making a phonographic record, and many others have synthesized

/Reed cut off ••.

$

of ...... ". ' ""': ..... '• ."•;, •'2'/• • Auxilia• • S Whistle

Leather Nostril X

Section through Resonator and Reed

Fro. 12. Wheatstone's speaking machine.

London & Westminster Review 28 (1837); The Scientific Papers of Sir Charles Wheatstone (1879), pp. 348-367. Proc. British Ass. Adv. Sci. Notices (1835), p. 14.

•8C. M. Gabriel, "Machine padant de M. Faber," J. de Physique 8, 274-5 (1879). F. Techmer, Phonetik (1870). F. Techmer, "Naturwiss. analyse und synthese der hrrbaren sprache," Int. Zeits. f. allg. Sprachwiss. 1, 69-170 (1884).

Helmholm, Sensations of Tone (1875), translated by Ellis. R. Koenig, Qudques Experiences d'A½oustiqu, (1882). D.C. Miller, Science of Musical Sounds (1916). C. Stumpf, Die Spraddaute (1926). W. H. Preece and A. Stroh, "On the synthetic examination of vowel sounds," Proc. Roy. Soc. London 28, 358 (1879).


SPEAKING X ACHINE OF WOLFGANG VO'q KEMPELE'q 165

Fla. 13. Faber's

speech organ.

vowels. The art of electronic music is closely related. A recently built 100-element tone synthesize• a has been provided with variable build-up and decay controls so that it can better simulate the tone characteristic of

various musical instruments, including, of course, the voice.

Paget made devices of plasticene, rubber, etc. for producing individually almost every consonant and vowel sound as discussed in detail. in his book? Wagner 26 built a vowel-copying electrical circuit to control the amount of power in the region of the fundamental frequency and in each of four formant frequency regions.

Some partial synthesis devices make use of the human mouth but supply a substitute energy source for the vocal cords. Thus, the artificial laryr•x of R. R. Riesz

u H. Fletcher, "Demonstration lecture introducing the new tone synthesizer," Am. J. Phys. LI, 215-25 (1946).

• Paget, Human Speech (1930). sa K. •. Wagner, "Ein neues elektrisches Sprechgerat zur

Nachbildung der menschlichen Vokale," Abhandl. d. Preuss. •.kad d. Wissenschaft. (1936L

referred to t• supplies a vibrating reed for vocal cords that have been removed or cannot be used satisfac-

torily. Wright •-• developed a "Sonovox" which uses other sound sources than the vocal cord tone for the basic

power. In particular, music or other material is obtained in electrical form from a phonograph record and con-

14. Stewart's electrical synthesizer for simple speech sounds. Courtesy of A'alure.

'• G. Wright, Electronics 13, 67 (August, 1940).



FIO. 15. Schematic of the yoder, electrical circuit for producing speech with manual controls. Courtesy of Journal of the Franklin Institute.

verted to electrical waves which energize a sort of bone conduction receiver to transmit the sound through carti-

ß lage of the larynx into the throat. Silent speaking modu- lates such power into sound patterns giving the effect of sound produced from other than vocal cord tones. The modulated output waves from the mouth at low l•vd are then picked up and amplified to produce unusual voice effects. Similar effects are producible electrically

Fro. 16. Schematic of the vocoder, automatic electrical speech synthesizer. Courtesy of Journal of the Acoustical Society of America.

with the vocoder mentioned later. Professor Firestone •a built and on November 3, 1940 demonstrated before -the Acoustical Society apparatus for projecting sounds from an electrical organ or other electrical source into the mouth whereupon mouthing gave a modulation to produce multi-voice singing and similar effects.

Stewart 29 first set up an all-electrical network as in Fig. 14 for making some of'the speech sounds. In 1939 an all-electrical speech mechanism known as the Voder aø (from the key letters of VOice DEmonstratoR) shown in principle in Fig. 15 was demonstrated by skilled trained operators at the New York and San Francisco World's Fairs. Figure 16 shows funct.ionally a corre- sponding device known as the vocoder • (from VOice CODER) having an electrical speech synthesizer similar to that of the yoder but making use of control currents from electrically analyzed speech for automatically operating the synthesizer instead of using manual controls.

28 F. A. Firestone, "Artificial larynx for speaking and choral singing by one person," J. Acous. Soc. Am. 11, 357, 376 (1940).

•9 J. Q. Stewart, "An electrical analogue of the vocal cords," Nature 110, 311 (1922).

a* Dudley, Riesz, and Watldns, "A synthetic spea•er," J. Franklin Inst. 227, 739 (1939).

an H. Dudley, "Remaking Speech," I- Acous. Soc. Am. I1, 169 (1939).


the speaking machine of wolfgang von kempelen (homer dudley and t. h. tarnoczy)

Documents