multimodalfactbullishAudiovisual expression dynamics carry identity-discriminative information complementary to static appearance in face recognitionComputer Vision28 Jul 2026http://arxiv.org/abs/2607.21347v1