Oh My Aachen Head - Arvin, Air Force develop neural network to interpret sound

When it comes to automated sound-quality testing, perhaps two heads are better than one.

In recent years, the automotive industry has relied on the Aachen Head, a microphone-equipped mannequin from Head Acoustics in Germany, to record the many sounds that flow from the tailpipe when a car is running.

But it was up to the human ears of trained sound specialists to analyze the source of those sounds and find out exactly why, for instance, a Neon sounds different than a Viper.

This year, the Aachen Head is likely to get a high-tech mate in the form of an advanced computer program specially design to quantify sound quality and respond to stimuli the way a person would.

The yet-to-be-named system is the result of a joint research project between automotive exhaust supplier Arvin Industries Inc. of Columbus, IN, and the U.S. Air Force Research Laboratory at Wright-Patterson AFB in Ohio.

An odd pairing, perhaps, but Arvin will use the system to develop automotive exhaust systems, while the Air Force will use it to formulate community noise guidelines for military bases, where soaring jets tend to upset neighbors.

The armed forces refer to it as a cooperative research and development agreement (CRDA), which allow for joint projects between the government and the private sector. In this case, both Arvin and the Air Force are contributing about $100,000 to the project.

The key to the system's success is the construction of mathematical algorithms and a neural network allowing the computer to essentially do what Arvin's sound specialists have done the old-fashioned way: by ear.

For Arvin, the tool could be especially useful in quickly tailoring an exhaust system to meet a particular customer's demands.

"Car companies come to us . . . wanting a specific sound for a specific car. They want a marketing image for a certain car," says John Grace, Arvin's vice president of systems and technology, who initiated the joint project and had worked with the Air Force at a previous employer. "Take the Viper - it does not want to sound like a purring kitten."

But Mr. Grace emphasizes that the project is still in the research mode. "Who knows where it can go? But in the process, we will learn a lot about sound characteristics and what's the difference that makes the difference and how do you identify it," he says. "Can I quantify to some level the characteristics of the sound so I may impact my design quicker?"

The new system would also reduce the subjective nature of sound analysis. The so-called "golden ears" in Arvin's research labs have been known to disagree on sound analysis from time to time.

"This is an aid to help them (sound experts) organize their thoughts, like a doctor doing a diagnosis," Mr. Grace says. "Can he diagnose the patient faster? Can this tool help you get there faster?"

Still, the neural network is not meant to replace sound analysts. In fact, those experts are playing a key role in developing a standard vocabulary necessary for the computer's algorithms.

The two-year research project began in May, and the goal is to develop one automotive system and one aircraft system, with both relying on the same neural network.

Granted, cars and jets make completely different sounds, but the beauty of the neural network is that it doesn't matter. The sound qualities can be different, but the system for studying them should be virtually identical.

Within the next six months the automotive system should be able to interpret up to six sound qualities, and that system will be considerably expanded within a year.

The research team will then set out to develop the aircraft system, which should be completed within a matter of months.

"From the Air Force, we look at aircraft engine sounds and what communities might find objectionable in a sound," says Bart Elias, who leads the Research Lab's Noise Effects Branch project. "When we fly over we get complaints occasionally. We want to know specifically what caused those complaints. Was it a low rumble or a high-pitched squeal?"

By law, the Air Force must address any type of noise it produces when planning to add runways at an air base. Mr. Elias says he is not aware of any particular base grappling with severe noise problems or planning to add runways.