Sound representation in higher language areas during language generation