Generate audio from omni-modalities in a single model.
Convert voices using audio samples
Convert or reconstruct voice recordings
