Has anyone tied this technique into deepfake analysis/synthesis?