Reddit post claims a new Scenema Audio release including model weights and inference code for zero-shot expressive voice cloning/speech generation, but the underlying paper/tooling details are not accessible from the snippet alone.