By the way, I would love a service where I could listen to HN. Play the first title and then wait for me to hit the space bar to play the next. Highlight the one I just heard, so when I'm interested int a submission, I have it right in front of me.
- Section headers are read just like regular body text, which is confusing
- Not great with reading numbers. Narrating a year is very different from narrating a monetary value.
- Parenthetical statements are read just like regular body text
It’s just missing some UX improvements now, mainly the ability to continue playing across multiple articles. It’s also not smart enough to skip boilerplate junk that appears on top and bottom of every newsletter, which is easy to skip over visually but a pain when read out loud.
A while ago, Amazon tried adding TTS to kindle and got immediately slammed by the publishers, so they withdrew it. However, with the current state of AI, I think users will start to demand it as a standard feature of anything they’re reading. Not only can it read at human-like capability, it can do so in any voice theoretically. Soon you’ll also be able to pause it and ask questions about what it just read or to summarize the rest of the article and so on.
But if you're looking for tech that can read it as well as a human can in terms of turning meaning into prosody -- all of the vocal inflections that we use and hear without even realizing it, the things that make it sound human and appropriate to the text -- that's hard to say. (Especially when you consider that most people are bad at audiobook narration, as it's a real skill.)
It might be very soon, if the kind of meaning embedded in LLM's turns out to be perfect for the job. Or it might be very far away, if it turns out that it requires building a kind of accurate model of human emotion and feeling and how that maps to vocal qualities, that has nothing to do with how LLM's work.
> Still, computer-generated voices have long struggled to hold the attention of listeners for long periods of time, and to overcome the ‘uncanny valley’ effect of synthetic human speech. Human intonation and inflection are notoriously difficult to predict and replicate.
I listen to a lot of audiobooks and I definitely notice and get annoyed when I don't like the narrator. I think I could maybe get used to this for one book but I would avoid it if possible, and I would really hate it if every book I listened to used this exact same style.
I would say this is probably already good enough for short non-fiction articles, but not there yet for fiction. It is genuinely confusing (for me at least) when an audiobook narrator reads a phrase with the "wrong" emotion, or doesn't vocally distinguish between characters.
It's perfectly fine for Siri responses or reading back a dictated text message. Not for listening to a whole book though.
Good modern audiobooks are a true performance even when the narrator isn't doing a bunch of highly distinct voices. You can't replace that with AI any more than you can replace a great voice actor.
https://www.kdpcommunity.com/s/article/Invite-Only-KDP-Beta-...
> Authors can set a list price between $3.99 and $14.99 and will receive a 40% royalty
Do they mean you get 40% of the work you created and they ran their model on? Oh lord. Let me run this through a TTS myself and publish it like that. Not sure if they even allow that though.
Audible's market share is like 60%, so you loose out on a lot if you don't publish there.
Before the acquisition, sonatics tech looked pretty cool. https://techcrunch.com/2022/06/13/spotify-is-acquiring-sonan...
that being said-- NOT A FAN. i want books read by human people who are good at reading to me. I dont want teddy ruxpin, i want my dad!
A colleague was saying to me that they would like to be able to have an ebook to read, but also to have it read out loud when driving or walking.
I think you need to set the Kindle app to continuous scrolling (in which case it won't work with older books), then you trigger the iOS screenreading shortcut, and off it goes.