The sensor is reportedly able to achieve far-distant and accurate voice recognition by mimicking the basilar membrane of the human cochlea, which can detect resonant sound using ~15,000 hair cell channels. Such flexible resonant acoustic sensors, say the researchers, promise to be an essential component for intuitive human-machine interaction (HMI) in the future voice user interface (VUI).

In their work, the researchers fabricated the mobile-sized acoustic sensor by adopting ultrathin piezoelectric membranes with high sensitivity. Simulation studies showed that the ultrathin polymer underneath inorganic piezoelectric thin film can broaden the resonant bandwidth to cover the entire voice frequency range using seven channels.

Based on this theory, say the researchers, they successfully demonstrated the miniaturized acoustic sensor mounted in a commercial smartphones and AI speakers for machine learning-based biometric authentication and voice processing. The resonant mobile acoustic sensor has superior sensitivity and multi-channel signals compared to conventional condenser microphones with a single channel, and it has shown highly accurate and far-distant speaker identification with a small amount of voice training data.

The error rate of speaker identification was significantly reduced by 56% (with 150 training datasets) and 75% (with 2,800 training datasets) compared to that of a MEMS condenser device.

“Google has been targeting the ‘Wolverine Project’ on far-distant voice separation from multi-users for next-generation AI user interfaces,” says Professor Keon Jae Lee, Dept. of Materials Science and Engineering, KAIST. “I expect that our multi-channel resonant acoustic sensor with abundant voice information is the best fit for this application.”

In 2018, Lee presented the first concept of a flexible piezoelectric acoustic sensor, inspired by the fact that humans can accurately detect far-distant voices using a multi-resonant trapezoidal membrane with 20,000 hair cells. However, previous acoustic sensors could not be integrated into commercial products like smartphones and AI speakers due to their large device size.

Lee has established a startup company – Fronics Inc. – in Korea and U.S. to commercialize this flexible acoustic sensor, which has been miniaturized for embedding into smartphones, and says the first commercial prototype is ready for accurate and far-distant voice detection.

For more, see “Biomimetic and flexible piezoelectric mobile acoustic sensors with multiresonant ultrathin structures for machine learning biometrics.”

Related articles:
3D-printed ‘cocktail party listener’ picks out voices in a crowd
Researchers unlock microphones to inaudible audio
Optical human-to-machine communication sensors promise ‘near-perfect’ speech recognition
Self-powered acoustic sensor boosts AI recognition


Linked Articles
eeNews Europe