CORNELL (US)—Researchers have developed a new method to help computers accurately simulate the sound of water dripping and splashing—and a range of other contact noises—by calculating how objects vibrate in the real world and how those vibrations generate acoustic waves.
The work by Doug James, associate professor of computer science, and his colleagues at Cornell University is the first step in a broader research program on sound synthesis supported by the Human Centered Computing Program of the National Science Foundation.
In computer-animated movies, sound can be added after the fact from recordings, but as virtual worlds grow increasingly interactive and immersive, the researchers point out, sounds will need to be generated automatically to fit events that can’t be predicted in advance. Recordings can be cued in, but can be repetitive and not always well matched to what’s happening.
“We have no way to efficiently compute the sounds of water splashing, paper crumpling, hands clapping, wind in trees or a wine glass dropped onto the floor,” the researchers say.
The algorithm the team devised can be applied to help simulate sounds made by objects in contact, like a bin of Legos, the noisy vibrations of thin shells, trash cans or cymbals, and the sounds of breaking glass and the clatter of the resulting debris.
All the simulations are physics-based and can be used in design, just as visual simulation is now, James says. “You can tell what it’s going to sound like before you build it,” he explains.
James and graduate student Changxi Zheng report that most of the sounds of water are created by tiny air bubbles that form as water pours and splashes. Moving water traps air bubbles on the scale of a millimeter or so. Surface tension contracts the bubbles, compressing the air inside until it pushes back and expands the bubble. The repeated expansion and contraction over milliseconds generates vibrations in the water that eventually make its surface vibrate, acting like a loudspeaker to create sound waves in the air.
The simulation method developed by the Cornell researchers figures out where the bubbles would be and how they’re moving, computes the expected vibrations, and finally the sounds they would produce. The simulation is done on a highly parallel computer, with each processor computing the effects of multiple bubbles. The researchers have fine-tuned the results by comparing their simulations with real water sounds.
The current methods still require hours of offline computing time, and work best on compact sound sources, the researchers say, but further development should make possible the real-time performance needed for interactive virtual environments and deal with larger sound sources such as swimming pools or perhaps even Niagara Falls.
The research was supported in part by an NSF Faculty Early Career Award to James, and by the Alfred P. Sloan Foundation, Pixar, Intel, and Autodesk.
Cornell University news: www.news.cornell.edu