Sensory Adds Suite of New Biometric Voice Features to TrulySecure Speaker Verification
Sensory, a Silicon Valley-based company focused on improving the user experience and security of consumer electronics through embedded AI technologies, announced the next-generation of its low-resource, noise-robust, TrulySecure Speaker Verification (TSSV) SDK. TSSV 2.0 boasts a variety of new embedded voice biometric authentication features and accuracy improvements.
TSSV 2.0 is a modular biometric software component providing speaker verification and identification. These biometric features run embedded and can be added to any on-device for profile switching, access security, and tracking conversations with personal assistants based on who is speaking. At its core, TSSV encapsulates novel advances in front-end noise suppression, robust feature extraction, as well as spectral and temporal speaker modeling to offer extremely accurate and secure voice biometric enrollment and authentication of users, even in noisy environments. The technology enables maximum privacy protections by running on-device and it supports multiple users with adaptation, allowing it to become more accurate and secure over time. TSSV can also be fused with Sensory's face biometrics, allowing for quick and seamless recognition and authentication of users via face or voice (convenience mode), or can be configured to require both face and voice confirmation for greater user security (TrulySecure mode).
TSSV provides device makers and app developers the flexibility to enable voice biometric security in three different modes, text independent, text-dependent fixed passphrase, and text-dependent user-defined passphrase.
New Features and Improvements for TSSV 2.0
- Accuracy improvements in feature extraction, reverb modeling, noise robustness/SNR have led to a decrease in equal error rates (EER) of more than 25% from TSSV 1.0.
- Wake word functionality for OS based systems. Wake-word creation is automated and fast once appropriate data is collected.
- Text independent speaker verification can track users by their voice without specific passphrase requirements.
- Compliant user training is no longer required to build voice models. Devices can with TSSV 2.0 can now listen to usage and build and adapt models without formal training.
- Prevents spoof attacks by presenting an interactive challenge sequence that must be performed to verify that the input is from an actual user and not a recording.
- Improved SDK - Documentation significantly expanded and improved. The SDK provides numerous code examples and now offers support for Python.
Sensory's TrulySecure Speaker Verification supports all languages and dialects.