Artificial intelligence device and 3D agency generating method thereof

An artificial intelligence device comprising:

• a memory configured to store training datasets for a target persona;
• and
• a processor configured to generate a three-dimensional agency corresponding to the target persona by using the training dataset for the target persona,
• wherein the processor is configured to:
• acquire a two-dimensional image including the target persona;
• pre-process the two-dimensional image to extract a training dataset including voice information, face information, and motion information corresponding to the target persona;
• input the training dataset to an agency generation model to learn one or more motion features of the target persona;
• and
• upon receiving user input data for generating the three-dimensional agency, inputting the user input data to a pre-trained agency generation model to generate the three-dimensional agency, the generated three-dimensional agency expressing the one or more motion features of the target persona,
• wherein the processor is further configured to:
• when learning the one or more motion features of the target persona, input a dataset including an identification code, voice information, face information, motion information, text information, and predetermined motion feature information corresponding to the target persona to the agency generation model;
• learn a voice of the target persona based on the identification code and the text information;
• learn a face motion of the target persona based on the identification code, the voice information, and the face information;
• learn three-dimensional rotation parameters of joints corresponding to the target persona based on the identification code, the voice information, and the motion information;
• and
• learn a unique specific motion of the target persona based on the identification code and the predetermined motion feature information,
• wherein the processor is further configured to:
• when learning a unique specific motion of the target persona, use the identification code to identify the target persona;
• and
• learn the unique specific motion of the target persona by controlling a weight so that the weight is lowered for a specific motion having a negative meaning and is increased for a specific motion having a positive meaning, based on the predetermined motion feature information of the target persona.