The BRUEL project concerns the evaluation/certification of voice identification systems in the face of adversarial attacks. Automatic speaker recognition systems are vulnerable not only to artificially generated speech, but also to other forms of attacks such as voice identity conversion and replay. The artifacts created during the creation or manipulation of these fraudulent attacks are the marks left in the signal by the text-to-speech algorithms, allowing the original real voice to be distinguished from a spoofed voice.

Under these conditions, the detection of impersonation requires the evaluation of impersonation countermeasures together with speaker recognition systems. The BRUEL project intends to propose the first evaluation/certification methodology for voice identification systems based on a Common Criteria approach.

IRCAM Team : Sound Analysis-Synthesis