ArmorVox is the only voice biometrics technology that can detect synthetic speech, digital voice and playback attacks.

Home / News & Media / ArmorVox is the only voice biometrics technology that can detect synthetic speech, digital voice and playback attacks.
 
ArmorVox is the only voice biometrics technology that can detect synthetic speech, digital voice and playback attacks.

Recently, voice imitation technology has been on the news for being able to create digital voice from a few minutes of audio sample. Lyrebird is one such start-up that has done a demo of this technology – and claims to be able to create digital voice from existing audio files.

While the technology is impressive and has numerous applications, this can be misused and is highly vulnerable to hacking threats – specifically for businesses that have implemented voice biometrics and those that are considering doing so.

At Auraya, we take security seriously and are constantly on the lookout for potential threats to our technology capability. The R&D team innovates for future potential risks and designs solutions that are robust and reliable, helping organizations provide unparalleled security for their customers.

ArmorVox is the next generation in voice biometrics technology – because of the security capabilities that differentiate it from other voice biometric offerings in the market. Here we summarize the features of the technology that defeat synthetic speech, digital voice and playback attacks.

 

Replay Attack Detection

The replay attack detection feature of ArmorVox matches the acoustic attributes of a voice sample against previous samples. If the match is too close, the sample is flagged as a potential replay of a previously presented voice sample.

Traditional text-to-speech synthesisers recreate the same synthetic voice sample each time to the same text – a replay attack ensures that you can differentiate between recordings or synthetic speech and natural voice.

 

Random Challenge Response

The Random Challenge Response is a standard feature of the ArmorVox voice biometric system. It generates a random challenge, such a random digit sequence, which the customer must repeat. To verify successfully, the customer is scored against repeating the right set of digits and his voice. This process overcomes regular pre-recorded playback attacks.

 

UBM Fit

Non-speech sounds such as enrolment and verification of tones, musical instruments, noise and synthetic voice can be potential threats to customer security and privacy.

UBM Fit checks the acoustic characteristics of a voice sample against the UBM of the voice biometric system.  The UBM is a special voice print model which is an amalgam of all the voices contained within the system.  If the acoustic attributes of the voice sample presented to the system generates a low “UBM fit” score, this indicates that the voice sample is not a voice.

UBM fit is also used to detect synthetic voice.  Given that synthetic voice contains acoustic artefacts that are not found in natural voice (hence the synthetic quality to voice synthesis); then UBM fit will return lower UBM fit scores for synthetic voice compared to natural voice.

Prior to voice enrolment, a UBM fit is performed to confirm that the voice sample presented for enrolment and verification is representative of natural voice and is not a non-voice or synthetic sounds.

 

Voice QA Model Pick Technology

ArmorVox Voice QA has “trainable” Voice QA “Model Pick” technology using Voice Signal Classification Models that the client application can utilise to provide a comprehensive signal analysis function and synthetic voice detection capability. 

This includes gender detection, and Voice Signal Classification Models for speech/non-speech detection; synthetic voice detection; age, accent/language and sentiment detection are currently in the works. 

ArmorVox can be designed to implement new Voice QA detectors to meet new and emerging voice biometric security threats. For example, if a new synthetic voice generator emerges as a security threat, a new Voice QA Model can be created using examples of the new synthetic voice which detects this specific voice quality and flags it as a security threat in the application. 

The ArmorVox Advantage

AURAYA SYSTEMS

Auraya is a world leader in biometric voice verification technology and empowers people and organizations to interact and engage with security and convenience. As a specialist voice biometric technology developer, we have a track record of delivering unparalleled security performance that is simple to deploy, integrate and maintain whilst delivering the most delightful customer experience.

At Auraya, we believe that the way you implement voice biometrics can make all the difference in realizing business value. We have spent decades gaining unique, real-world insights into key business drivers and customer expectations for voice biometrics and innovate for evolving market opportunities. Our proprietary voice biometrics technology ArmorVox helps support real-time fraud reduction with the Advanced Security Suite.

Connect on LinkedIn
Send us an e-mail
RSS
Share on Google+
https://aurayasystems.com/2017/12/08/armorvox-detects-synthetic-speech-digital-voice-playback-attacks">
Follow on Facebook

Leave a Reply

Your email address will not be published. Required fields are marked *