Verisoul Docs


Verisoul uses voice recognition to help ensure the user is a real, live and unique person

Process Overview

  1. 1.
    First, Verisoul captures a voice recording via the client’s microphone
    • Verisoul’s frontend SDK records the user speaking the displayed phrase in their preferred language
    • Saying the exact phrase is loosely enforced given the variability in the user’s grammar and voice proclivities (accents, conjugation preference, etc.)
  2. 2.
    Next, the recording is encrypted and converted to a VoicePrint
    • The voice recording data is actually doubly encrypted (similar to Verisoul’s facial biometrics). The frontend SDK encrypts the data prior to SSL/TLS standard encryption.
    • The recording is then converted to a VoicePrint and the original recording is discarded (not stored)
      • A VoicePrint is a non-reversible array of bytes that encodes unique voice attributes like tonality.
  3. 3.
    The VoicePrint is then validated for liveness and authenticity. Even though the VoicePrint is non-reversible it still contains the necessary information for Verisoul to run quality checks
    • Verisoul checks to ensure the print is not a spoof. This includes validating the speech was not recorded or artificially generated.
  4. 4.
    Verisoul stores NO speech, recording or audio data. Only an encrypted and irreversible VoicePrint is stored for comparison.
    • Data science models analyze VoicePrint in combination with other off-chain and on-chain data to determine user uniqueness.


Note: that the more VoicePrint a user has, the more accurately Verisoul is able to identify that user. If your use case requires a very high standard of authentication we recommend onboarding each user with more than one phrase however it can depend on the context greatly. Get in touch with us to learn more about our authentication services.