SAMO: Speaker Attractor Multi-Center One-Class Learning for Voice Anti-Spoofing