Is Aging Voice A Concern For Voice Biometric Authentication?
Pryathna Sankaranarayanan | April 5, 2018 | 4 minutes
Some industry participants have raised concerns that voice biometrics solutions may be subject to failure due to changes in an individual’s voice caused by ageing. These changes are claimed to have a negative effect on accuracy of voice authentication. After all, the voice you have as a teenager is different from the voice you have as an adult – can the system recognize your voice as you age?
Some claims have been made that the technology cant keep track of vocal changes that occur over time. If the system doesn’t track changes then eventually the systems rejects the true speaker and becomes less secure.
Many voice biometric vendors have attempted to address this concern by using historical voice recordings of high profile public figures over a period; such as President Obama, Arnold Schwarzenegger and Morgan Freeman etc., to show how they can identify a speaker despite changes in voice over time.
This approach, however, fails to consider real-world production deployments where voices and audio systems are changing in different ways. Different microphones, for example, capture your voice differently based on audio received from these devices.
There are many advantages of choosing a voice biometric technology that has been designed with machine learning capability – the more an individual uses the technology for voice verification, the better it gets at learning the different characteristics that make up the users voice. This overcomes concerns related to an ageing voice or changing devices and audio systems.
ArmorVox is Auraya’s proprietary technology designed with advanced machine learning algorithms. For every customer interaction, it updates the speaker’s voiceprint, along with the speaker’s background model and threshold settings to ensure that the system remains optimised for that speaker even as they age. Learning and adapting to customer voice is a continuous process. ArmorVox supports subtle changes in a customer’s voice by continuously learning the underlying vocal characteristics of the customer each time they verify.
13 February 2020
06 February 2020
15 January 2020