OpenMOSS / OmniVAE

A unified audio-video variational autoencoder with cross-modal alignment for joint reconstruction and generation