Page 45 - Special Issue
P. 45

 Figure 4. The “talking heads” of the Abbé Mical, exhibited in Paris in 1783. The dialogue spoken by the two heads is written on the curtain hiding the mechanism (see text for actual French): “1st Head: ‘The King gives peace to Europe’; 2nd Head: ‘Peace crowns the King with glory’; 1st Head: ‘And peace makes the people happy’; and 1st Head: ‘O adorable King, father of your people, their happiness shows Europe the glory of your throne.’” At the bottom of the display is the caption: “Talking Heads: A problem solved in mechanics that up to this day had been considered unsolvable, or at least very difficult. The Academy of Sciences has said in its report that these talking heads can throw the greatest light upon the mechanism of the vocal organ and on the mystery of speech. It added that this work was worthy of its approval by its novelty, by its importance, and by its execution.” Illustration reproduced from the Bibliothèque Nationale, Paris, France, with permission. Copyright Bibliothèque Nationale, Paris, Gallica, ark:/12148/btv1b8410437r.
This mechanism is remarkable in the history of speech syn- thesis for several original contributions. The Abbé Mical’s “talking heads” constituted the first automatic programmable speech synthesizer. They were the first synthesizer based on
concatenation of phonetic units and also the first synthesizer based on replicating the serial ordering of articulations seen in the human vocal tract, using a form of demisyllable synthesis. Finally, as the examples above show, they synthesized the first dialogue between two machines. Against all of these advances is the undeniable deficiency that they seem to have relied on coordinating and switching between multiple separate mecha- nisms to produce the illusion of speech instead of controlling a single unified model of the vocal tract, which partly accounts for the poor and unnatural sound quality.
Faber’s “Euphonia”
In the early nineteenth century, an Austrian mathematician and astronomer, Joseph Faber, came across Kempelen’s book and built a replica of the mechanical speaking machine for himself. Quickly realizing the problems involved in playing it by hand, he determined to improve on the original design by turning it into a keyboard instrument to automate the means of control. Like Kempelen’s machine, he used bellows and a vibrating reed as lungs and larynx to provide an airflow and sound source. However, he replaced the simple half-open mouth tube that Kempelen shaped by contortions of the hand with a fully configurable model of the whole vocal tract. By mounting six adjustable metal blocks back-to-back in a square box behind a pivoting tongue resting on a rotating jaw that terminated in a pair of moveable rubber lips, he was able to create tube shapes with front and back cavities that could be easily and directly related back to actual vocal tract configura- tions. As in a real vocal tract, the tongue, jaw, and lips moved to create simple opening and closing movements in the front cavity of the model, incorporating natural articulatory con- straints, whereas the blocks behind resemble a six-section area function that could be used to create a more detailed shaping of the back cavity. Following Kempelen, Faber also added a more realistic nasal cavity, inserted a rotating vane to inter- rupt the airflow for trills, and used a moveable lever to alter the effective length of the reed, raising and lowering the pitch or cutting off the reed vibration completely.
The real ingenuity of Faber’s talking machine, however, lay in the control mechanism. The jaw, tongue, lips, six blocks, nasal opening, vane, and reed were all controlled by a set of vertical rods such that the continuously adjustable heights of the rods entirely determined the configuration of the
vocal tract model at any point in time. Faber connected the rods to a keyboard with 14 keys and 3 pedals, using a grid of cross-patched levers to transform each single key or pedal depression into a predetermined setting for multiple rods. He
Spring 2020S, uSpmemciaerl I2s0s1u9e | Acoustics Today | 145 Reprinted from volume 15, issue 2

   43   44   45   46   47