A View from Dresden onto the History of Speech Communication
Part 5: Artificial vocal fold models – The investigation of phonation
The investigation of the larynx was (and is) one of the predominant topics in phonetic research. In the early times of experimental phonetics, mechanical models of the larynx or, at least, of the vocal folds have been utilized according to the paradigm of analysis-by-synthesis.
The first models used flat parallel elastic membranes or other simple elements to simulate the function of the vocal folds (Fig. 1). However, the geometry of these models was rather different from that of real human vocal folds. A substantial progress was made by Franz Wethlo (1877 – 1960), who worked at the Berlin University as an educationalist and special pedagogue. He realized that the vocal folds should not be modelled by flat parallel membranes, but that the three-dimensional shape of the vocal folds should be taken into account. Hence, he proposed a three-dimensional model, which was formed by two elastic cushions (Fig. 2). The cushions were filled with pressurized air, the pressure of which could be varied for experimental purposes. In particular, the air pressure in the cushion pipes was varied to adjust the tension of the vocal folds. The whole model was known as “Polsterpfeife” (cushion pipe). Wethlo described it in 1913.
The historical collection (HAPS) at the TU Dresden owns several cushion pipes from Wethlo in different sizes, modelling male, female, and children’s voices. A team from the TU Dresden repeated Wethlo’s experiments with his original equipment in 2004. Therefore, the cushion pipes were connected to a historical “vocal tract model”. This vocal tract model was actually a stack of wooden plates with holes of different diameters to model the varying cross-sectional area of the vocal tract between the glottis and the lips (Fig. 3). This “configurable” vocal tract model came to the HAPS collections from the Institute of Phonetics in Cologne. The artificial vocal folds were used to excite vocal tract configurations for the vowels /a/, /i/ and /u/, but listening experiments showed that these artificial vowels were rather difficult to discriminate.
Today, there is renewed interest in mechanical models of the vocal folds. Such models can be used in physical 3d robotic models of the speech apparatus (e. g., the Waseda talker series of talking robots: http://www.takanishi.mech.waseda.ac.jp/top/research/voice/), to evaluate the accuracy of low-dimensional digital vocal fold models (e. g., http://scitation.aip.org/content/asa/journal/jasa/121/1/10.1121/1.2384846) or to examine pathological voice production.
Rüdiger Hoffmann & Peter Birkholz