Engineers Eavesdrop On "Aeroacoustics" Of Human Voice

"The main interest is to try to predict the consequences of surgery and maybe plan the operation to minimize the effects on voice," says Luc Mongeau, an associate professor of mechanical engineering at Purdue University.

He and associate professor Steven Frankel, with funding from the National Institutes of Health, have created plastic and mathematical models to better analyze and recreate the voice production process, which depends on turbulent air flow through the glottis, an opening into the voice box.

The researchers are trying to predict the aeroacoustics, or the aerodynamic sound, produced by air flow. Findings may not only help to preserve a person's voice but also may help engineers figure out how to better synthesize and characterize the voice for robotics and voice-recognition purposes.

The Purdue engineers have detailed some recent findings about the work in a paper appearing in the March issue of the journal Physics of Fluids, published by the American Institute of Physics. The paper focuses on the aerodynamics behind "impulsively started" jets of air that are central to human speech.

"Impulsive is when you build up pressure and then you release it all of a sudden," says Mongeau.

The voice process begins when the lungs exert air pressure and the vocal cords open, releasing successive, pulsing jets of air. Each jet of air is attached to a leading vortex, which resembles a smoke ring, that eventually detaches from the jet. The time it takes for the ring to detach from the jet – about one thousandth of a second – is critical to the formation of speech.

"We looked at that process with a magnifying glass in a big computer simulation to try to understand that type of flow better," Mongeau says. "What we want to know is, how much jet development you have during that period of time, and is that sufficient for a single vortex to form and detach, or would it stay attached until the formation of another one?

"Think about smoking a cigarette and making smoke rings. If you make them very slowly, the rings have time to go away, and you can watch them dissipate. But you could also puff them in close succession, and that's when you get what I call a vortex train, one vortex following another, and it looks like a caterpillar."

The computational results discussed in the paper supported previous work by other researchers, and they also revealed something new: Each individual jet becomes unstable and forms tiny eddies that influence the ring's detachment. The numerous eddies also affect the eventual qualities of the human voice.

In addition to their work involving computational simulations, the Purdue researchers have designed an artificial larynx, the structure in the trachea that houses the vocal cords. As air flows through the model, its rubbery walls are rapidly adjusted by small rods to simulate how the tissue responds during speech.

"What we are doing right now is extending this work one step further and actually taking into account the motion of the walls, so that the vocal cords are now moving," says Frankel. "The air pushes on the walls and then the walls spring back, pushing on the air. So, there is an interaction between the two."

Eventually, the modeling research will have to extend beyond the larynx, if engineers are to fully understand the physiology of speech. The strategy is different than a more conventional approach to speech synthesis, which ignores human physiology. The physiological approach strives to model speech production by taking into account the positions of the vocal cords, tongue, lips and other "articulators" involved in speech, Mongeau says.

Such a technique enables scientists to "develop a sort of code book so that for a given sentence, you track the motion of all the different articulatory parameters," he says. "And then from the articulatory parameters, you generate the speech output."