The language of sounds unheard: Exploring musical timbre semantics of large language models