
The clatter of a dropped trash can and the crash of a cymbal ā both easily recognisable sounds.
Thatās why computer games or CGI movies that feature such noises use samples recorded from life, not generated by software as the graphics have been. It would take weeks of intense computing to synthesise the sound of a single cymbal clash.
New methods developed at Cornell University in Ithaca, New York, change that and could make objects in games and movies sound more like those in real life.
Advertisement
Thin shells
The sounds the new method targets are those made by āthin shellsā: objects like clashing cymbals or falling plastic bottles, whose thickness is much smaller than their other dimensions.
They typically produce loud sounds ā think of hail falling on a tin roof ā because the thin shell readily vibrates and radiates sound into the surrounding space. But they defy the acoustic computer models normally used to reproduce virtual sounds in games and films, says at Cornell.
Those models assume that the dominant vibrations in an object act independently, an approach that works well for a chime, say. But the sound of a thin shell like a clashing cymbal results from dominant oscillations coupling with and influencing one another.
āIf you include those neglected non-linear contributions, you start recovering some of the richer sounds,ā James says. But although simulating complex interacting vibrations is vital to the design of everything from aircraft to nanoscale structures, āthe standard thin-shell vibration models can be prohibitively slow for sound renderingā, he says.
Bottle smash
āIt took us a couple of weeks to synthesise a bottle impact sound [a few seconds long] using standard engineering methods,ā says James, who with colleagues developed a model that can achieve similar results in just a few hours.
Thatās possible thanks to carefully targeted simplification. Instead of calculating the oscillating forces across the thin shellās entire surface, the forces at work at only a relatively small number of points are considered and used to approximate the forces across the surface.
Removing that computational bottleneck led to a dramatic drop in the computation time necessary to produce the sounds. Using conventional methods, a 5-second clip took 19 days to generate on a cluster of 16 high-powered dual quad-core computers. The same job took just 90 minutes with the simplified approach.
Umbrella bats
āWe use linear sound models all over the place because theyāre cheap and simple, but by devising efficient numerical methods you can capture these non-linear effects and get far more interesting sounds,ā James says.
, a computer scientist and seasoned sound engineer based in Bournemouth in the UK, thinks Jamesās teamās techniques, and their work on generating water sounds earlier this year, is āinteresting and valuableā.
It isnāt yet suitable for real-time game audio, he says, but its time will come: āNo doubt by the time Hollywood and the game industry wake up to whatās going on, computers will be fast enough to run [these] implementations in real time, so this work is amazing for the future,ā he says.
However, Farnell thinks itās moot whether ārealisticā sounds will necessarily play well with audiences. āI think we have a misplaced obsession with ārealismā,ā he says. For instance, movie-goers have become so accustomed to hearing the sound of a shaking umbrella when a bat flies across the screen ā a well-established sound effect ā that āa recording of a real bat is weak and disappointing,ā he says.
Jamesās team will present their work at the conference in Yokohama, Japan, in December.