Speech Technology Magazine

 

Sensory Releases Embedded Speaker Verification SDK

New speaker verification component of the TrulySecure platform can be integrated into devices as either text-dependent or text-independent.
Posted Jun 30, 2015
Page1 of 1
Bookmark and Share

Sensory has released an a la carte software development kit for the embedded speaker verification component of its TrulySecure technology platform.

By embedding Sensory's new speaker verification component into a device, manufacturers will be able to deliver new experiences in which a user's voice can be used to switch profiles, access specific data, or track conversations with personal digital assistants.

Sensory's new speaker verification technology boasts up to 96 percent success at recognizing the correct user and 99.99 percent success at rejecting imposters. For less than ideal environments with more ambient noise, the technology still offers an equal error rate of 2 percent or lower.

"We are always exploring new avenues for improving the security and user experience of consumer electronics," said Todd Mozer, CEO of Sensory, in a statement. "We pioneered low-power text-dependent speaker verification in our TrulyHandsfree offering. With TrulySecure, we have pushed the boundary on performance and text-independent usage, and, because of the strong interest we have received, we decided to take the speaker verification component of TrulySecure and wrap it up into its own SDK for our customers to begin using."

Sensory's TrulySecure speaker verification component is designed to automatically identify and authenticate users based on their voice. The technology offers support for multiple enrolled users and can adapt to each user's voice, becoming more accurate over time. This technology uses advanced noise-robust signal processing to deliver excellent performance and accuracy in a broad range of environments, including outdoors and in moderate-noise. Speaker verification can be combined as a modular component with TrulyHandsfree for ultra-low power, always-listening wake-up with ultra secure biometric authentication by using two independent verification approaches. User enrollment only takes a few seconds and as little as a few spoken words. The technology is designed to constantly adapt to the users' voices, allowing its performance to perpetually improve. It is language-independent, so language-specific models are not required and accents do not adversely effect the accuracy.

"What sets this technology apart, other than its incredible accuracy, is the fact that it can be incorporated into devices in either text-dependent or text-independent modes. So, depending on how it's integrated by the manufacturer, a passphrase may be pre-determined or left to the end user to choose, or the application may identify the user from their unique characteristics from random speech," Mozer added.

SDKs for the embedded speaker verification component of TrulySecure are available for Android, Mac OS, Linux, and Windows applications, with support for other leading platforms coming soon. Jibo, the world's first social robot, is the first customer to license this technology and will use it to track users in a room. Sensory plans to integrate this new technology as part of the company's TrulySecure biometrics technology offering.


Page1 of 1