+7 (342) 218-38-68

Metamaterials Single-Out One Voice

Back

LAKE WALES, Fla.—Researchers at Duke University (Durham, North Carolina) have invented a metamaterial-based voice-locator system that can pick out an individual's voice even from a crowded, noisy room.

The researchers' novel technique to solve the classic coctail party problem uses a metamaterial rotunda structure with different depth holes sunk into its pie-shaped unit cells. Sound waves going over them make a sound using the same principle as blowing over the top of a pop bottle, enabling the embedded electronics to distinguish the direction from which a voice is coming.

The voice-zoom option on video cameras and smartphones today can beam-steer to a particular voice, using multiple microphones and sophisticated algorithms to pick out one voice from many. However, the metamaterial rotunda's unit cells employ a single microphone for acoustic direction location plus sophisticated algorithms to do a superior job, according to Duke.

"Metamaterials are used here to modulate sound waves directly without converting them to electronic signals first," doctoral candidate, Abel Xie, working with professor Steven Cummer. "Each metamaterial unit cell changes the sound's amplitude and phase at a specific frequency. The properties of these unit cells are randomized so each sector (or slot) possess a unique modulation for sound wave passing through it. Thus sound waves from different directions will be encoded differently by metamaterials which we can separate with our algorithms."

Metamaterial rotunda locates specific voices even in a crowded noisy room for easy voice-command recognition.
(Source: Duke University)
 

Instead of just the loudness of the sound coming through each sector or slot in the metamaterial rotunda imparts both amplitude and phase modulations that are different for each one — a signature that the algorithms recognize.

"Each sector or slot is an encoding channel and adds a unique 'signature' to the sound passing through it," Xie told us. "Technically speaking, the 'signatures' are highly frequency dependent amplitude and phase modulations that are different from channel to channel. The algorithm knows what the modulations are like so it can separate the mixed sound to output high fidelity sound."

The biggest engineering hurdle to commercialization is the large size of the prototype, but the Duke researchers claim that they can downsize the design to work in smartphones, by changing the materials and high-fidelity requirements.

Testing the accuracy--over 96.7 percent--of the metamaterial rotunda in locating distinct voices within a 360 degree periphery.
(Source: Duke University)
 

"Firstly, we are looking into other types of acoustic metamaterials—instead of air-filled cavities used in this work—that can create similar modulations within much smaller volumes. Secondly, the miniaturization will be performed for specific applications, for example, for a speech recognition function in a game console, we probably wouldn't need as much sensing capacity and full-angle range as demonstrated in the work, so the size can be made much smaller," Xie told us.

The technology will enable smartphone users to speak into their phones from a distance, even though they are in a noisy environment. Other applications could be to improve the fidelity in voice-command electronics, medical sensing devices, ultrasound imagers, hearing aids and cochlear implants according to the researchers.

The large size of the metamaterial rotunda will be shrunk to smartphone size before commercialization according to the researchers.
(Source: Duke University)

Source: www.eetimes.com