MDG Group introduced a new version of the multimodal biometric platform

Posted by

The STC Group of Companies has introduced a new version of the multimodal biometric authentication platform. This was announced at the St. Petersburg International Economic Forum by the CEO of the MDG group of companies Dmitry Dyrmovsky. The project was implemented with the support of the Russian Foundation for the Development of Information Technologies.

VoiceKey.Platform is a platform that combines products and services for customer identification and fraud prevention in remote services based on voice and facial biometrics. The main benefits of biometrics are increased security and improved customer experience. An updated version of the biometric platform can be in demand in banks, retail, MFC, telecom, development and other industries.

How the platform works: when a call is made to the contact center, at the moment of starting a conversation with the operator, a user check is launched – a voice model is created in real time, compared with the standard, and monitors the appearance of the voice of strangers, including previously compromised ones. The whole process takes a few seconds. The result of identity verification by voice appears on the monitor screen of the contact center operator. With the introduction of such a solution, there is no need to remember the number of a card, passport, code word or other identification information: to receive personal information on your accounts or transactions, you only need a voice. At the same time, the system will instantly recognize the artificial voice and report that the verification has not been passed. The solution does not depend on the language the client speaks. All this makes biometrics not only more convenient than traditional methods of identity verification, but also much more reliable.

Thanks to the support of RFRIT, it was possible to implement a number of global platform updates in a short time: improve the user interface, develop technologies for voice and facial anti-spoofing (protection of biometrics from hacking).

Voice biometrics examines unique voice characteristics and “recognizes” a person by voice, which is more reliable than, for example, a request for personal data – a request to name the last digits of a passport or the last completed banking transaction. To pass voice biometrics, scammers can play a recording of a person’s voice or take a technologically more complicated path – fake the desired voice using Deepfake technology. To detect such actions, Antispoofing technology is used – protection against hacking.

Facial biometrics in Russia expects rapid growth

Import substitution

The R&D team of the STC group within the framework of the project developed methods (algorithms) for detecting various types of attacks on voice biometrics. In 2021, the R&D achievements of the MDGs in this area were recognized at the global level: the team won first place in a prestigious international scientific competition. The competition was close to real conditions: in some scenarios, the data was recorded in telephony with various encoding effects, and in other scenarios, the data was processed by various codecs that compress audio with losses. Thus, in the competition, the teams dealt with the specifics of the signal from a variety of channels that attackers can use.

Three scenarios were the subject of the study: attack detection based on replay. In this type of attack, the scammer obtains a recording of the user’s voice in some way, and then plays that recording when attempting to hack; detection of attacks based on voice synthesis or transformation. With such an attack, fraudsters can use arbitrary user speech samples (for example, from audio files available on the Internet) and, using a cloning or voice transformation system, obtain synthesized speech to hack a biometric system; detection of Deepfake attacks, where it was proposed to distinguish real voices from artificially created ones in difficult acoustic conditions.

The STC group was able to achieve high results in protecting biometrics through the use of new architectures of neural networks, approaches to their training, as well as new non-standard ways of augmenting training data (making changes to the data in order to increase their variability and obtain a more stable algorithm).

Source link

Leave a Reply

Your email address will not be published.