Zero-resource Speech Translation and Recognition with LLMs