Page 43 - Special Issue
P. 43

 Figure 2. Kratzenstein’s five vowel tubes for the vowel sounds a (two cross sections in Figs. 9, 10), e (Fig. 11), i (Fig. 12), o (Fig. 13), and u (Fig. 14), excited by a free reed (Figs. 5-8). Kratzenstein remarks that the passage from o (Fig. 13) to u (Fig. 14) is achieved by the stricture of the upper cavity. Reproduced from Kratzenstein (1780).
machine, as summarized in his famous book (Kempelen, 1791). After working on various pipe models like Kratzenstein, Kempelen eventually returned to analogies between music and speech. Inspired by a rustic bagpipe, he adopted the familiar model of wooden bellows, counterbalanced by a weight, to blow air through a vibrating ivory-and-leather reed mounted inside a box and venting through a flared gutta-percha tube, respectively simulating lungs, larynx, and mouth (Figure 3). In the process of developing his model, he thought carefully about the relationship between the detailed articulations of all the different parts of the vocal tract and the phonetic contrasts that he found to be important in different languages, realizing that many speech sounds can be compared and discriminated based on voicing, aspiration, frication, or nasality as well as the shape of the vocal tract. Accordingly, to the basic model he added nostril tubes that could be opened
or closed; a means of damping the vibration of the reed to cut off voicing; an extra smaller side bellows that would inflate and then rapidly deflate as the mouth opening was closed off and released to create air puffs for plosives; two side tubes of differ- ent lengths bypassing the reed for fricatives; and, finally, a metal wire that could be pushed onto the reed to create a rattle resem- bling a trill. The bellows were pressed with the right elbow, and the levers and openings on the top of the main box were oper- ated with the right hand to control the secondary modifications, while the left hand was partially inserted into the mouth tube and manipulated empirically to create the primary articulation of the sound. Modern reconstructions have shown that, in the hands of a skilled operator, unrestricted whole sentences can be produced intelligibly on demand, but the perceptual quality is far from natural, as everyone remarked at the time (Liénard,
1967; Brackhane, 2011).
These first two vocal tract models have been described in detail elsewhere many times (e.g., Chapuis and Gélis, 1928; Dudley and TarnÓczy, 1950) and are the most well-known but perhaps the least interesting of all the mechanical speaking machines. Kratzenstein’s tubes reproduce isolated vowel sounds acous- tically but fail to accurately simulate even the geometry of the vocal tract, let alone the underlying physics. Kempelen’s speaking machine captures many of the physical mechanisms responsible for sound production in the vocal tract, albeit crudely, but totally sidesteps many of the crucial issues of articulatory timing and control by reducing the model to a pas- sive musical instrument that needs to be harnessed back to the actions of a human musician. The shaping of the mouth tube, for example, is largely provided by the operator’s hand and can only be learned through a great deal of practice and listening.
The next two models ultimately resolved these problems.
The “Talking Heads” of the Abbé Mical
An almost exact contemporary of Kempelen, the Abbé Mical was an impecunious cleric, the younger son of a wealthy family from the Dauphiné in France who ran away from the church to pursue mechanics. From 1776 to 1785, he exhib- ited across Paris first one, then two, carved wooden “talking heads” that produced not only single speech sounds and syl- lables but also whole sentences and even an entire dialogue (Figure 4). On his request, his invention was examined by the Académie des Sciences, which appointed a committee of notable scientists, including Vicq d’Azyr, Laplace, and Lavoisier, to produce a report (de Milli et al., 1784; Chapuis and Gélis, 1928; Hémardinquer, 1961).
Spring 2020S, uSpmemciaerl I2s0s1u9e | Acoustics Today | 143 Reprinted from volume 15, issue 2

   41   42   43   44   45