Voice ID On-device - Embedded Secure Authentication Solution

Easy, Embedded and Secure Voice Biometric Authentication for Devices and Applications

Sensory’s TrulySecure Speaker Verification (TSSV) technology is a language independent, highly secure yet convenient, on device verification of a person’s voice or passphrase. The underlying technology leverages Sensory’s deep strengths in speech processing, noise mitigation, speech feature extraction and machine learning. TSSV offers ease-of-mind specifications:

NMo special hardware is required. TSSV uses standard microphones on phones, tablets and PCs. Additionally, all processing is done on-device so personal data remains secure. TrulySecure Speaker Verification features a fast and simple to use enrollment in which users repeat a passphrase (either predetermined or user-defined) three or more times. TSSV provides maximum security against unauthorized attempts to break in, while ensuring the highest verification rates for the user.

It is robust to environmental challenges such as noise and therefore it

works in real-life situations.

The TrulySecure Speaker Verification also features SoundID, a recognizer capable of identifying a variety of environmental sounds, including glass breaking, babies crying, dogs barking, home security alarms, smoke/CO alarms and low battery warnings, doorbells, knocking, snoring and coughing.

TSSV supports all major operating systems, offering nearly limitless implementation flexibility. Additionally, Sensory can customize TSSV to match the exact needs of its customers, enabling only the sound profiles required for specific use cases.

Modalities, Features, & Specifications

Sensory’s TrulySecure Speaker Verification (TSSV) SDK provides an efficient on-device technology that supports development of audio-focused products incorporating voice biometrics, wake-words and sound-identification.

The core voice biometric capability consists of a voice enrollment and authentication process.

The SDK supports three basic types of authentication scenarios, Fixed -Phrase, Text-Dependent, and Text-Independent.

Text-dependent authentication requires the user to repeat a selected passphrase during enrollment. Generally 3 or more repetitions are required with 4 repetitions being recommended for most applications.

The table below summarizes MIPS requirements for the case of non-target speech and audio regions as this represents the vast majority of audio the engine processes in normal use-cases. Additional specifications for fixed phrase and Sound ID models are available upon request. Speeds are typical for Raspberry Pi 3 model B.