Phonexia launches deep neural network-powered voice biometric engine
Phonexia has announced the launch of its Deep Embeddings voice biometrics engine, which integrates deep neural networks (DNN) to match voices to voiceprints, and which the company says is the first commercially available voice biometrics engine to use this machine learning capability.
The technology identifies unique features in each voice with a discriminative training model to create voiceprints twice as fast, with 2.4 times better accuracy, and with one quarter of the memory requirement of Phonexia’s previous engine.
“The technical benefits -accuracy, speed, and reduced memory use – from transitioning completely to deep neural networks have exceeded our expectations,” stated Petr Schwarz, Phonexia CTO.
The Deep Embeddings engine is part of the modular Phonexia Speech Platform, which in addition to the speaker identification and verification provided by Deep Embeddings, provides speech-to-text, keyword spotting, language identification, gender identification, and language estimation within a scalable platform which is easily integrated with into other solutions, according to the company.
“In addition to making biometric adoption easier for traditional clients, the reduced memory requirements will accelerate adoption of speaker identification into new segments such as 4.0 devices, IoT, and devices with no permanent connection to the Internet,” explained Schwarz.
The Czech Republic-based Phonexia will look to make inroads in the growing voice biometrics market, which Nuance Director of Product Strategy for Voice Biometrics Brett Beranek told Biometric Update in a recent interview is benefiting from the evidence rolling in from early adopters of its effectiveness.