Team of Speech Technology Center Beat Market Leaders in the International Scientific Competition on Biometrics Protection from Hackers



The system created by developers and scientists from Speech Technology Center made minimum mistakes defining replay attacks at the recent Automatic Speaker Verification Spoofing and Countermeasures Challenge.

The competition is held once in two years at the largest speech technology conference Interspeech. It is organized by scientists from the European universities; its participants are also teams from large companies, research institutes and universities. So, for example Samsung and Pindrop, which are acknowledged leaders in the world market of voice biometrics, presented their solutions too.

Each time organizers offer fighting attacks of a certain type which are imitated as realistic as possible. To decide the winner the number of true and false positives of biometric systems presented by competitors is assessed. This time participants were offered fighting audio replay attacks. At the attack of such a type a fraudster makes a record of user’s pronunciation of a passphrase, and then replays this record at the system tampering attempt in order to perpetrate the user’s identity.

The solution of Speech Technology Center has shown the best result in the competition with equal error rate of 6.73%. The system occupied the second place was almost twice as worse — 12,34%.