It still has much room for improvement

It still has much room for improvement

The model you see here is actually von Kempelen's third attempt at building a speech synthesizer. As Joseph mentions, the first wasn't much more than a bagpipe reed and a kitchen bellows, and the second was more like an organ, with an individual pipe and key for each phoneme. The third design was an attempt to more closely emulate the human mouth, nose and throat, and was one of the first machines to successfully do so.

However rude it may appear on first glance, this final design is incredibly clever; you can see how it works in this interactive flash site from the Kempelen Farkas Speech Research Laboratory in Budapest. The Institute of Media Archeology in Austria also has their own replica, and their site offers sound and video clips. Here's one more replica for good measure.

Von Kempelen usually made it pronounce words and short phrases in Latin, Italian, or French rather than German, because the machine had difficulty with harsh consonants. Despite twenty years of work and continual adjustments to improve its inflection and pitch control, its voice remained very crude (and rather creepy). Still, it was an amazing effort for one man, and von Kempelen's contributions to the field are still noted today.

Speech simulation has taken quite a different turn these days -- scientists have largely given up on replicating human speech through physical means because it's just too difficult. Instead, research facilities have turned to creating huge libraries of individual sounds in sentence context.

The original speaking machine still exists, and can be viewed by the public in the Musical Instruments section of the The Deutsches Museum in Munich.