TMT: Tri-Modal Translation between Speech, Image, and Text by Processing Different Modalities as Different Languages