Fooling Speaker Recognition Systems with Voice ImpersonatorsMarie Donlon | November 15, 2017
As mobile devices with applications functioning via voice command become more and more common, researchers from the University of Eastern Finland have found that skilled voice impersonators can often bypass sophisticated speaker recognition systems, potentially posing serious security issues.
Often used to authenticate the user of an application or for public safety purposes, voice recognition is likely more vulnerable to live voice impersonators that can make voice modifications than to the technically-generated voice attacks.
Generally perpetrated using technical measures such as voice conversion, replay and speech synthesis, the scientific community has been more successful developing countermeasures against these attacks. However, the countermeasures do not easily detect human voice disguise and impersonation.
Analyzing the speech of two imitators mimicking a few well-known Finnish public figures, the study found that the impersonators successfully “fooled” both listeners and automatic systems with their impersonations.
The study is published in Speech Communication.