Audio-driven unified speak-listen facial motion generation