New biometric sensor can discern 9 voices in a package small enough for a phone
Granularity is being brought to acoustic sensors for phones. A team of South Korean researchers have created small, flexible piezoelectric membranes with enough resonant bandwidth to cover the human voice frequency range.
The highly sensitive and ultra-thin sensor is modeled after the inner ear’s cochlea, and can be integrated with phones and digital assistant speakers. There, they would support machine learning-based biometric authentication and voice processing.
The work was done at Kaist, or the Korea Advanced Institute of Science and Technology. Researchers say the new voice recognition sensor is 70 percent smaller and 20 percent more sensitive than previous versions.
Nine registered voices can be identified by the prototype, which Kaist says is ready for commercial applications. The prototype can recognize a registered person based on one training word and one testing word.
Beyond being ready to commercial integration, the Kaist researchers are pitching the sensor for use in a Google project, called Wolverine, seeking to separate the voices of people distant from the listening device.
Media reports indicate that Wolverine wants to give people not just good hearing, but hearing that surpasses what is experienced by people with excellent hearing.