MotorMouth detail
MotorMouth, detail of the head and throat
MotorMouth is an acoustic speech synthesizer with just one voice pipe. It is a mechanical version of the human mouth with moving lips, teeth, tongue, tip of the tongue, larynx and a blower which serves as the lungs.

These components and the motors and mechanisms which drive them are clearly visible. It is programmed with a few sentences and can pronounce individual vowels, semi-vowels, nasals, labials and fricatives. It can speak with varying pitch which it demonstrates when it says "I love you" and it can count from 1 to 10 in English and German.



Audio: counting from 1 to 10 and "I love you"



MotorMouth on Vimeo in a new window
In Vimeo you will have to scroll down and click on the full screen symbol, bottom right.



The project was a reinterpretation of the human vocal tract (below) as a piece of functioning machinery. It took me a couple of years - most of this time was spent making very small adjustments, both to the mechanism and to the software, listening to the results and then making some more very small adjustments. This painstaking process seems to have stilled my appetite for speaking machines ... although it is excruciatingly tempting to start on a new and improved version incorporating the experience gained.


a section through the human mouth
the vocal tract

The vocal "cords" are two soft flaps of flesh on opposite sides of the air pipe leading from the lungs. Normally thay are separated - for breathing. When we speak they move closer together, almost touching, and vibrate in the wind flow from the lungs, slapping together to produce the basic unfiltered voice sound. (They can also be tensioned to raise the pitch). The tongue changes this sound by moving backwards, forwards, upwards and downwards inside the mouth cavity, so changing its resonant frequencies. The varying apertures formed by the lips also change the mouth resonances.


the first sketch
initial sketch

MotorMouth is an attempt to reproduce some of the features of the vocal tract using 8 motors. The blade of the tongue can be moved forewards, backwards, up and down. The lips, teeth and nose valve can be opened and closed.



MotorMouth general arrangement drawing
general arrangement drawing

As in the case of its predecessor, the Talking Machine, the voiced sounds are produced by a vibrating reed. Unlike humans, MotorMouth has two air supplies: one air valve to vibrate the voice reed and a bypass air valve for unvoiced sounds. These unvoiced sounds, like F, S, T, etc. require much more air than could easily pass through the reed itself. If I was going to do this project again today, I would make the vocal cords in a much more realistic way, probably using a very soft rubber, such as Septon.


drawing of a human and MotorMouth speaking the vowels
the vowels

A comparison of how a human being and MotorMouth speak the vowels.


MotorMouth
MotorMouth (1996-1999)

8 stepping motors, blower, micro-computer, interface, wooden case. 86cm high.
At the top, the head of the machine and, below, the stepping motors that drive it via cranks and bowden cables. The motors are controlled by a built-in Z80 board - the 1990s equivalent of an Arduino. At the time this was a reasonably fast and compact system and I had all the hard- and software available and ready to go. The keypad is used for launching the speech programs and for changing parameters such as overall pitch.
Seven motors can be seen below the head. The eighth motor that turns the tongue is on the rear side of the machine and, like the blower, not visible here.


Technical details

Photo of seven motors, Z80 board and interface.
The eight motors are controlled by a micro processor board from Ing.Büro Kanis and driven by an interface that was very kindly designed and built for me by Dipl. Ing Gottfried Müller. It features eight SAA1027 stepper driver chips that take care of the step patterns. I had previously used the same hardware and motors and similar software to drive a kinetic sculpture: Straight Eight (1993). The reference mark sensors of the motors share their outputs with the 7 outputs of the keypad which is disabled while the motors are running.


Preliminary working drawings

Drawing of a section through the tongue and motor
section through tongue and motor

The section shows how the tip of the tongue, the "blade", is moved up and down by a Bowden cable.

The surfaces of the blade are covered with PTFE coated fabric to reduce friction.

The drawing shows a roller bearing for the tongue. In the final design this was replaced by a much wider phosphor bronze bush bearing taking up the entire 20 mm width of the base plate. The clearances between the tongue, chin and base plate are set to about 0.1 mm,
Drawing of the mouth components
Detail photo of mouth

The tongue, chin and base plate were cut from solid aluminium with an industrial CNC milling machine by Michael Buchmann. To make it lighter and easier to move fast, the tongue was hollowed out to within 2 or 3 mm from the outside surfaces and the cavity was filled with a hard polystyrene foam and then sealed. The lips, teeth and the blade of the tongue are made of wood. The blade was also drilled out to lighten it.


Software


Data:
How to say "ten"

The eight 1's and 0's under the line vnpbtlaT select which motors will be moved next:

voice-valve
nose-valve
pitch
blade of the tongue
lips
teeth
air-valve
Tongue

The following lines show the position that each motor should go to and its speed. In the case of "ten" there are three stages: T - EH - N. Other symbols such as Tee are shorthand for standard routines, in this case a very quick release of air to say "T". Nose 0, at the end, closes the nose valve.
MotorMouth program
Program:
Interpreting the data

The Z80 microprocessor is programmed with about 46 pages of heavily annotated assembly code which boils down to about 3 pages of machine code that the Z80 chip can actually understand. About half of this is data: the speeds and positions of the motors that enable the machine to speak, (see above).

The other half, the program, seen on the left, is about interpreting that data and driving the motors.

Assembly language: "LD A, (HL)" etc. etc. is rather dry but very fast; at the time just fast enough. Today I would use a more comfortable language and a normal, faster computer.


What they said:

A successful and ingenious masterpiece by Martin Riches: an artificial speech mechanism using motors, valves, a rotor and a blower that can say How are you? or I love you! at the touch of a button. Its speech may sound a trifle hollow, the voice of a survivor perhaps, but it is fascinating to see how an artificial tongue and lips can produce sounds similar to those of a human being without recourse to any form of recording. It is also somewhat uncanny: the ability to speak is generally taken as a definitive sign of the development of human intelligence but here we bear witness to the articulations of an automaton.

    Tom Mustroph   Exhibition review, Neues Deutschland, Berlin 16.06.2002

Martin Riches' MotorMouth is not concerned with the shortcomings of human attempts to represent Nature but is rather an audacious attempt at its direct reproduction. Having closely studied the ways of human speech he has transformed them directly into mechanical technology and has succeeded in making a machine that can speak. The machine produces its programmed words without a mechanical accent. Art and Nature are unmistakably reconciled.

    Anne Thiem   Exhibition review, Berliner Morgenpost 26.08.2002

I found a friendly technical review of this webpage on Hackaday 2017/02/08 reminiscing about Z80 assembler programming. I learnt assembler language in the 1980s on a Sinclair ZX81 and continued using it after most sensible people had graduated to C++ or Python.

    Steven Dufresne   MotorMouth for Future Artificial Humans, Hackaday 02.08.2017


I was also pleased to find the following text by Heinrich Heine, written in 1834:

The story is told of an English inventor who, having devised the most ingenious machines, finally hit on the idea of constructing a human being. At last he achieved this ambition and the work of his hands was able to behave and comport itself just like a human. It even bore within its leathern bosom something quite like human feelings - not unlike the normal feelings of an Englishman. It could communicate its sensations in articulate sounds and the noises of the internal wheels, ratchets and screws which could then be heard gave it an authentic English accent. In short, this automaton was a Perfect Gentleman and lacked nothing to transform it into a real human being ... except for a soul.

But this the English mechanic was unable to give him and the poor creature, well aware of this deficiency, now tormented his creator day and night with the request to give him a soul. These requests, repeated ever more urgently, became so unbearable that the artist finally took flight from his own creation. But the automaton followed post-haste, pursued him to the continent, caught up with him occasionally and then snarled and grunted at him: Give me a soul!

Heinrich Heine
On the History of Religion and
Philosophy in Germany
(1834)
Third Book

drawing of an android saying A

Thanks, acknowledgements and history

I am especially grateful to Michael Buchmann who programmed and operated the industrial CNC milling machine to cut the aluminium components. I am no less grateful to Dipl. Ing. Gottfried Müller who designed, built and tested the extremely compact interface that controls both this and an earlier kinetic work.

The project was supported by a grant from the Department of Cultural Affairs Berlin in 1996. It was first presented - at a very early stage - by Freunde Guter Musik Berlin on 17 November 1996. (It shared the programme with a performance by Emmett Williams.)

During the production period I received encouraging communications from Dr. Christophe d,Alessandro of LIMSI-CNRS, Paris, who is both an organist and a developer of digital speech synthesizers; and also from Dr. Hideyuki Sawada of the Kagawa University who built the Kagawa Talker (also known as Motormouth!)   I was visited by Dr. Masaaki Honda of Waseda University together with a colleague, while they were attending a phonetics conference in Berlin and showed them my MotorMouth. In 2007 I visited the Takanishi Laboratory at Waseda University, home of the Waseda Talker Project where I was very pleased to shake the hand of Dr. Atsuo Takanishi.

MotorMouth is now in the collection of the Berlinische Galerie, Berlin's municipal museum. They have kindly lent it for a number of exhibitions, including the conference and concert The Art of Voice Synthesis at Het Orgelpark, organised by the University of Amsterdam, 11 - 13 May 2016 and the exhibition Two Measures of Time, with Max Eastley, at the Stadtgalerie Saarbrücken, 16 Sept 2016 - 8 Jan 2017.

Finally, I would like to share a mnemonic for medical students that I found in the course of my anatomy research:
  The lingual nerve describes a curve around the hyoglossus.
  "Well, I'm fucked!" said Wharton's duct,
  "The blighter's double-crossed us!"

Wharton's duct is situated under the front of the tongue. It supplies saliva.
MotorMouth has a lubrication hole at approximately the same location.

to speaking machines