Vulnerability of Automatic Identity Recognition to Audio-Visual Deepfakes