Voice Biometrics

 Voice Recognition

Voice Authentication is a new technology and increasingly used tool in the process of user authentication. Not only shortens the user verification time by more than four times, but also increases the safety of this process.

Contrary to the password, which can get into unauthorized hands, the voice is unique for every human being. This ensures that the users’ data are adequately protected and can be accessed in a quick and comfortable way. Conventional, widely used PIN or password based authorization method is not only uncomfortable for users, but also costly.

Our solution is a voice biometrics producer with successful international implementations in fintech, banking, insurance, telco, public.

Voice recognition solution is used for authentication, login, personalization, fraud detection on mobile apps, helplines, call centers.

We believe that voice is the most convenient and natural way of communication. We want to scale our business more in such innovative areas like IoT and blockchain.

How it works?

 Voice Biometrics Technology is based on analysis of the distinctive patterns contained in the human voice, impossible to reproduce.

A voiceprint is made of user’s voice patterns that are stored in the form of mathematical models.

Our Solution is resistant to spoofing. It detects whether the sample is an original or a re-played recording.

Each time a speaker is being authorized, recording is verified and matched with all previously collected data guidelines.

Our Solution is robust to perceived voice changes. It works even when you have a cold or a sore throat. What it listens to in your voice, is not what people hear in it.

 Text Independent System

The main function of the system is to verify the identity of users and to monitor unauthorized access attempts – also alarm activation, if necessary, automatic blocking of unauthorized user or prompting suggestion to add suspect to the Blacklist.

Deployment of our  biometric voice Text-Independent system does not require from the Call Centre customer more than mere speech. The system relies on the safety, convenience and voice communication channels speed of use.

System features:

  • Detection of access  attempts by impersonation
  • Verification of the correctness and consistency of speech samples used for training and testing the System
  • Monitoring of the signal quality for speech training and testing
  • Intelligent adaptation of the system to new voice patterns
  • Constant efficiency of the solution regardless of the language used
  • Support for multiple voice communication channels (GSM, PSTN, VoIP, WEB)
  • Compatibile with automated dialog  systems  that use automatic speech recognition (ASR) to confirm the identity by using additional criteria (eg. user number, social security number)
  • Scalability according to the number of registered users and the number of simultaneous channels
  • Integration with existing ICT solutions (call centers, IVR systems, infolines)  via dedicated API
  • Compliant with all application security standards

Solution Flow Blacklist

In voice-driven customer service channels like contact-centers, web and telephone helplines, the following important aspects must be addressed: customer privacy, customer data security and customer experience. Focusing on customer experience often leads to the weakening of the security of the solution. Knowledge-based security solutions reduces customer experience, increases interaction handling time and, finally, are prone to fraud attacks. Knowledge-based security is prone to human factor-caused flaws. Nowadays, professional fraudsters can collect customer-sensitive information using web resources. Such data is then used for fraud attacks in voice-oriented customer communication channels. Criminals perform well planned assaults based on identity theft and customer impersonation. According to the latest reports, more than 25% of successful identity theft worldwide takes place in contact-center channel. In most cases, such attacks are indistinguishable from legitimate customer conversation and occurs repetitively.

Our Solution Flow text-independent passive customer authentication is the new layer of possible security solutions. It reduces threats of fraud attacks based on identity theft and impersonation. At the same time, effortless and transparent  our solution Flow authentication increases customer experience and reduces necessary authentication time.

Our Solution Blacklist is an additional, spoofing and fraud protection solution which adds another layer of security to any voice-based customer channel. It allows for detection of repeating fraudsters using their voice-biometric traits. With VoicePIN Blacklist all previously noticed criminals can be detected during any future fraud and spoofing attacks.

How it works: Any previous known recorded fraud attempt can be used to create a fraudster voiceprint entry in the Blacklist using one or more recording of the criminal’s voice. To start build the Blacklist using known attacks, because most fraudsters are returning persons. The blacklist can contain several entries of different fraudsters which are concurrently assessed and monitored providing the maximum possible level of biometric security. Once the particular fraudster or fraudsters are added to the Blacklist, the fraud alarm will be raised anytime their voices are spotted during any customer interaction.

The most important benefits of Our Solution FLow Blacklist deployment are:

  • voice-based fraud level reduction by 80% on average,
  • increase in customer experience (CX), by possible elimination the need of additional, extensive security questioning or customer token requests in case of higher operation risk situations,
  • reduces fraud assistance and fraud handling costs, both in terms of financial and operational effort,
  • both Flow customer authentication and Blacklist are passive interaction-transparent solutions,
  • Blacklist adds a new layer of security without disturbing the legitimate customer experience,
  • Blacklist rise alerts in real-time during the conversation and reduces fraud-recovery efforts.

Key features of Our Solution Blacklist are:

  • transparent and passive customer voice monitoring,
  • false alarm rate below 0.1%,
  • more than 80% of fraud attacks detected,
  • easy to use administration and reporting web-based graphical user interface (GUI),
  • false alarm sensitivity calibration regardless the blacklist size,
  • detected fraudster identification (one-of-N) and identification confidence score output,
  • possible multiple independent blacklist sets within a single VoicePIN deployment,
  • possible white-list functionality configuration within the same deployment,
  • recommended from 1 up to 100 voiceprint entries in a single Blacklist (and up 200 or more entries if higher false alarm rates are acceptable),
  • straightforward REST API-integration with your existing risk and business logic management tools and infrastructure,
  • each blacklist entry can be created using one or multiple voice files if at least 40 seconds of net speech length is available,
  • off-line scan of archived recordings to detect any previous possible fraud activity.

Anti Fraud

Fraud Discovery system has the ability to create new patterns of voice for the needs of Blacklist based on database records from intrusion attempts (made, for example, on the basis of determinations agents), all of which have captured most frequently appearing voices.

Anti-Fraud is a set of utilities allowing for an efficient discovering of fraud attempts that may occur during the process of communication in Call Centers, infolines, video-chats, etc.

 Components:

  • Flow – passive identity verification that allows for matching the streamed voice with previously registered samples (e.g. during recorded phone call). Agent has access to real-time verification scoring and may react and direct the conversation so the Caller would be authenticated without any additional steps, such as security questions.
  • Blacklist – black list of users (available for Flow) and is capable of storing up to 100 speech pattern for each of communication channels. This component allows for quick detection of conversations where known, blacklisted voice of a fraud appears. Black list check occurs in real-time during the call.  New fraud voice patterns may be added manually by Administrator or by Agent.
  • Fraud Discovery – allows for creation of new voice samples for Blacklist according to recordings of fraud and spoof attempts (flagged as such by Agents). Voices that repeat most often are being registered. This functionality facilitates creation of the Blacklist by automatisation of the process of grouping the recordings and allows for most efficient fraud detection model creation. Analysis of historical recordings creates the suggestions for the list of serial fraudsters to be included on the Blacklist.

Text Dependent – Fixed Password System

Fixed Password – are designed to provide significantly efficient and secure biometric identity verification. Using fixed-phrase during the “training” process allows to uniquely link the voice biometric model with phonemes of the password phrase – thus the system, besides spectro-temporal voice features, is also sensitive to high level features such as intonation or accent. The enrollment, as well as verification, is much faster than in text-independent mode. In this type of systems, the common and fixed passphrase is used by all speakers for both enrollment and verification.

Fixed Password features:

  • Verification of the correctness and integrity of test and training samples,
  • Checking the quality of test and training samples,
  • Smart adaptation of the system to new user voice samples,
  • Detecting unauthorized access attempts by sample analysis,
  • High accuracy regardless of the language,
  • Support of several voice communication channels (GSM, PSTN, VoIP, Mobile, WEB),
  • Scalability according to the number of registered users and simultaneously used channels,
  • Integration with the existing computerized systems (hotlines, IVRs combined with ASR systems, conventional IVRs) via a dedicated API,
  • Application of valid application safety standards.

Playback Detection

Audio replay spoofing constitute a threat in biometric speaker verification systems. It occurs when a  previously recorded vocal passphrase is played back by an unprivileged person in order to gain access to someone else’s account.

Our voice biometrics solution solves this problem. Our system implements Playback Detection capabilities which protect each user from such recording-based spoofing in fixed-passphrase mode. Each utterance of the biometric passphrase is different from other utterances. Even two similar utterances from the single speaker are in reality slightly different from each other due to behavioural reasons. Playback Detection measures such variability and utilizes it to tell whether the provided sample is a new utterance, or one of the previously observed utterances of the particular speaker. Therefore, playback attack can be detected if the sample used for spoofing was intercepted using software (internal) or hardware (external microphone) eavesdropping during any previous verification of a legitimate speaker’s account.

Fixed-passphrase Playback Detection feature can easily be toggled using Our Solution web-based administration platform.