ASVspoof 2021 Dataset Review

The ASVspoof challenge is the premier benchmark for audio deepfake detection. The 2021 edition introduced new challenges, including deepfake speech in compressed codecs and physical access attacks.

Logical Access (LA)

This scenario focuses on synthetic speech (TTS) and voice conversion (VC) attacks transmitted over a telephone or VoIP channel.

Physical Access (PA)

This scenario involves replay attacks, where a recording of the target speaker is played back through a device in the same physical space as the microphone.

Speech Deepfake (DF)

A new track focused on detecting compressed deepfakes, simulating the scenario where fake audio is shared over social media platforms that apply lossy compression.