Making an AI audiobook of a classic? Check whose translation you're reading.

You find a classic novel, choose an AI voice and start making an audiobook to share. Before generating chapter one, I would check the translator named in that particular copy. A novel being in the public domain in the United States does not establish that a recent English translation is too. The US Copyright Office's registration help gives a plain example: an English translation of a public-domain Spanish novel, with the original text excluded from the claim and the translation included. Section 103 of US copyright law separately explains that protection for new material does not extend to the preexisting work. That cuts both ways. A publisher cannot reclaim the old text merely by adding new material, but the old text's status does not clear the new translation or introduction for reuse. I would ask the assistant to identify the exact edition, translator and source of the text, then help find a version whose reuse status can be checked for the country where I plan to publish. ‘The author died centuries ago’ is not enough. These sources explain the distinction; they do not decide whether a particular recording is lawful or whether an exception applies. Checking the copy first leaves more room for the enjoyable choices: which voice suits the story, and where to pause. It is a much smaller job than replacing the text after producing twenty chapters.