“The time needed for voice identification has been reduced from 10 to seven seconds, while enrollment time has been brought down from 35 to 20 seconds.”
Phonexia has launched the fourth generation of its Deep Embeddings biometric voice recognition platform, delivering significant improvements over the 3rd generation system.
Next Generation Voice Biometrics
Whereas the previous iteration of Deep Embeddings sported a 1.24 percent Equal Error Rate – making it “one of the most accurate on the market”, according to a Phonexia statement – the 4th generation platform brings it down to 0.96 percent. And when applied to existing client datasets, the system’s accuracy has doubled in some cases.
Deep Embeddings is also much faster now. The time needed for voice identification has been reduced from 10 to seven seconds, while enrollment time has been brought down from 35 to 20 seconds. And it can process audio recordings twenty times faster than real time, a fourfold increase compared to the 3rd generation.
The Deep Learning Behind Deep Embeddings
The upgrade comes after the breakthrough of the addition of machine learning to the Deep Embeddings upgrade delivered about a year ago. Now, Phonexia says its platform is “the world’s first commercially available voice biometrics engine based exclusively on deep neural networks”, which could further help to generate enthusiasm given the growing excitement around machine learning-driven biometrics, not to mention voice-based device interaction.
April 4, 2019 – by Alex Perala