Here is the way it used to be done before AI took over, when it was called speech synthesis. How do you know all of these programs were created ethically and responsibly and nobody ever used any pirated software or infringed any patents or copyright? You don't, some people maybe did, but it's probably okay these days.

You can also find reasonably responsibly trained machine learning models that are open and able to run locally. Nobody's going to promise there was never any part of it that was trained on data you wouldn't approve of. There's simply no way to tell. We don't have any system capable of identifying or defining who is consenting or what they are consenting to.

So, you really have to define this line yourself, and you have to understand that nothing's going to be done perfectly ethically without question. If you're willing to compromise a little without having unrealistic expectations, and live in the unfortunate reality that we find ourselves whether we like it or not, you can probably find some free, locally-hosted model that will do a much better and more capable job than any previously mentioned software solution, but you have to accept the risk that even then maybe it's not as ethically trained as it was claimed to be. We don't know, nobody can promise or verify that.

There is no ethical free lunch. If you want to avoid "AI" completely you can try to find and hire a real human being to read to you. Even that's hard, and it's certainly not immune to perpetuating some form of exploitation. We live in a world of exploitation. You can try your best to minimize the harm you do, but the philosophy gets complicated fast.

And to quote "The Good Place": this is why everyone hates moral philosophy professors.