Skip to content Skip to main navigation Skip to footer

Testing possibilities

The best option for a complete testing of Voice Verify is obtaining an evaluation license (provided by Phonexia) of the full solution and integrating it into a contact center. However, sometimes it is faster to use other available options, in case only the core voice biometric technology (without additional features) is to be tested or the integration needs to be done very quickly. All of these testing options can be used in a Proof of Concept, however integration of the evaluation package into a contact center is the preferred way as it can be turned into production instantly (from the technical point of view). Let’s take a look at all available possibilities.

1. Voice Verify DEMO

Voice Verify DEMO was created specifically for demonstration purposes. By using this demo you can get the feel for Voice Verify in a couple of minutes even without any programming or technical knowledge. Voice Verify DEMO can help in understanding the process of enrollment and verification by let’s you try our technologies on your own voice. For access credentials please contact your sales representative.

What can be tested via the DEMO?

  • Voice biometry technology
  • Enrollment (voiceprint creation)
  • Verification

Requirements

  • Web browser
  • Mobile phone to receive incoming calls
  • Username and password created by Phonexia to log in

Benefits

  • Mobile phone can be used for easy testing
  • No prior programming knowledge needed
  • Runs in cloud provided and maintained by Phonexia – no hardware needed for the voice biometry technology

Limitations

  • No additional features (calibration, logging, connection to a contact center,…)
  • Only live calls can be processed
  • Voice Verify Demo works with outbound calls
  • The accuracy measurements may not be fully transferable into a production environment (more information below)
  • 3 minutes call-duration limit
  • Maximum 10 calls per day per telephone number
  • Demo access requests must come from official company’s email address
  • Requests received from Gmail, Yahoo, Hotmail and other such domains are subjected to additional verification

2. Voice Verify evaluation package

The best way of testing Voice Verify is of course via the evaluation package of the complete Voice Verify solution. Voice Verify can be integrated into a contact center and tested out on the real production data. In case the integration into a contact center is not a straightforward task and the testing needs be done very quickly, Phonexia is able to provide a complete testing environment including the Voice Verify evaluation package, open source Asterisk IP PBX (https://www.asterisk.org/) and a guide on how to use them together with any kind of softphone application.

2.1 Voice Verify evaluation package tested with Asterisk

What can be tested via the evaluation package?

  • Voice biometry technology
  • Full voice biometry integration via RESTful API
  • Full set of features
  • Interaction with a provided open source IP PBX

Requirements

  • RESTful API knowledge
  • Experience with running virtual machines
  • VMware Workstation or ESXi hypervisors
  • Two virtual machines provided by Phonexia
    • Voice Verify virtual machine
    • Asterisk virtual machine
  • Softphone application (our recommendation is using Linphone https://www.linphone.org/)
    • requires PC/server with GUI or can be installed on mobile phone
  • On-premise hardware – minimum 4 CPU cores and 12 GB RAM
  • All components running in the same network with no closed ports, DHCP needs to be set-up correctly

Benefits

  • Very quick and easy integration with Asterisk IP PBX
    • No integration into a full contact center is needed
  • Supports both live calls and voice recordings
  • The complete RESTful API integration can be done
  • Full set of features available

Limitations

  • Runs on-premise only – proprietary hardware needed
  • All PC’s/cell phones/servers/virtual machines have to be connected to the same network
  • The accuracy measurements may not be fully transferable into a production environment (more information below)

2.2 Voice Verify evaluation package tested in a contact center

What can be tested via the evaluation package?

  • Voice biometry technology
  • Full voice biometry integration via RESTful API
  • Full set of features
  • Interaction with any contact center software

Requirements

  • RESTful API knowledge
  • Experience with running virtual machines
  • VMware Workstation or ESXi hypervisors
  • on-premise or private cloud hardware – minimum 4 CPU cores and 10GB RAM
  • Voice Verify virtual machine provided by Phonexia
  • Integration into a contact center software/PBX

Benefits

  • Supports voice recordings for enrollments and live calls for verification and enrollments
  • Complete integration into a contact center software
  • Full set of features available
  • Solution is fully prepared for production once the testing period ends
  • Accuracy measurements are transferable into a production environment
  • Runs both on-premise and in cloud

Limitations

  • Integration with PBX could pose a challenge

Results

The ultimate goal of voice biometric testing is finding the false acceptance and false rejection rates (for more information, see Speaker Identification) in verification attempts. Whether or not was a specific verification result correct has to be written down manually by the the person being verified or into a custom graphical user interface developed for this purpose. There is no graphical user interface provided by Phonexia for this testing evaluation.

Examples:

  • Person A tried to verify as Person A
    • Verification result is verified -> correct verification attempt has to be written down
    • Verification result is not verified -> incorrect verification attempt has to be written down
  • Person A tried to verify as Person B
    • Verification result is verified -> incorrect verification attempt has to be written down
    • Verification result is not verified -> correct verification attempt has to be written down

(the cases when not sure is returned have to be treated separately).

Transferability of accuracy measurements

The goal of voice biometric accuracy measurements is to find out how well will Voice Verify perform in terms of false acceptances and false rejections (for more information, see Speaker Identification). To ensure transferability of these measured results into the real production environment, the testing dataset has to match the production data perfectly, including:

  • Source (GSM, VoIP,…)
  • Microphone placement (close-field vs. far-field)
  • Audio quality
    • Formats
    • Codecs
    • Background noise
  • Geological locations
  • Age distribution
  • Style of speech
    • Monolog vs. dialog
    • Reading a text vs. live conversation

In some of the scenarios mentioned above, it is quite difficult to assure all of these requirements, that is the reason why the best option for accuracy testing is definitely in the production environment.