Voice recognition technology has been developing for around 70 years now. Nowadays, AI-fueled voice solutions can analyze human voice and derive lots of data from it, including emotional state, native region, and lots more.
These developments make voice an efficient and unique fingerprint that can be used for security reasons. Today we will talk about voice biometrics and what exactly modern AI can tell about you using just your voice.
Voiceprints for faster user identification
It turned out that like fingerprints, we all have unique voiceprints. Each human body is unique as well as the voice itself, which has lots of specific parameters that differ from person to person. These include pronunciation, emphasis, speed of speech, accent, timbre. Thus, voice can be used as an additional security token for more efficient authentication. Financial companies worldwide are actively using such technologies.
For example, the AI can automatically analyze the person’s voice calling the contact center in less than 20 seconds. While the person describes his or her problem, the system conducts automatic verification that reduces the time spent on this task by 66%.
Citibank is one of the companies that use such voice biometrics technology. The bank’s customers can enroll by recording their voices. These voiceprints are then stored in the database and used for identification every time they contact the bank.
The quality of voice biometrics is very high nowadays. For example, HSBC UK’s voice biometrics system blocked 2x more fraud attempts in 2019. We at Neuro.net also develop a voice biometrics technology based on Text Independent Engine with a typical error rate (ERR) of less than 2%.
Another popular use case of voice biometrics is mobile device user identity confirmation. For example, Android-based phones users can unlock their gadgets with voice using Google Assistant.
Another use case of voice biometrics is fraud prevention and detection. If the voice can be used to identify the legitimate customer and speed up the verification process, why can’t we use voice to detect fraudsters? Sure, we can, and businesses in multiple countries are already using AI-fueled voice biometrics solutions to do it.
Such systems usually whitelist voiceprints of regular users and blacklist voices associated with the confirmed fraud attempts. Then the system maps every caller’s voice with these white and blacklists to conduct authentication.
For the contact center usage, the system can update the CRM’s customer profile assigning the corresponding flags. For example, red is a recognized imposter, and green is for the verified customer. If there is not enough data for the system to decide on the caller’s security level, it can run additional verification by asking more custom questions.
This approach is wildly popular among financial companies. For example, Citibank and ANZ use voice biometrics to identify fraudsters and authenticate callers proactively at their call centers.
Using voice as an additional identification token is not new. For example, by the early 20th century, law enforcement departments in the US and Europe used vocal portraits as parts of criminal records.
Fighting internal threats
Dealing with fraud by actors attacking companies from the outside is essential, but insider threats pose even more significant risks for companies. This is obvious, as it is much easier to attack, say, a bank from the inside instead of trying to breach multiple security systems protecting its perimeter.
The cost of insider threats (related to credential theft) for organizations in 2020 is $2.79 million, which is enormous. Banks may also lead to reputational risks and regulatory investigations and massive fines.
This is why financial companies are now paying lots of attention to a proper employee or contractor’s assessment. Unsurprisingly, voice biometrics tech is on the rescue here as well. One of the ways of using such systems is measuring the risk in a specific voice. Here is how it works: a job applicant answers a set of questions. The AI can detect each answer’s stress level based on neurophysiological reactions and score the overall risk.
Companies with a distributed workforce must make sure the specific information can be viewed and altered only by trusted employees. Companies also implement voice-fueled employee validation for internal operations like help desk or call center contact requests to mitigate the risk of fraud. AI efficiently fights social engineering when cybercriminals mimic legitimate employees to access essential data or break into the corporate infrastructure.
Moreover, voice can also be used to detect accents and understand the person’s country of origin. For example, in the early 2000s, some countries, like Australia, used language tests for asylum seekers. When such people arrived in the country with no documents, the language tests allowed linguists to detect their country of origin with a shallow error rate.
So, today, if the job applicant or contractor claims he or she is from one country, but AI thinks their accent matches another one, this is an obvious red flag.
Voice biometrics is on the rise nowadays. Solutions like Neuro.net allow companies worldwide to increase the overall security level and prevent fraud, decrease risks of insider attacks, and implement more effective authentication systems for their customers.
The standard security workflow assumes the need to remember dozens of passwords, keywords, document numbers, or pins for the customer. This is hard, so people try to make their lives harder at the security cost (e.g., they pick simple passwords and use them in multiple accounts).
AI-fueled voice biometrics allows freeing customers from these complex tasks and significantly improves user experience – imagine a bank that is not asking pin codes for its customer for years but recognizes him by voice? Sounds surprising, but this is already a reality when innovative voice technology allows people to skip unnecessary interactions with the business and get to the point right away while keeping the security at the highest level possible.