This article is disparaging Project Gutenberg so badly that to me it almost seems to be trying to hurt the cause of the actors currently on strike.
Spreading FUD about text-to-speech technology that is not completely lifeless is a bit like blaming google translate or OCR programs from taking away jobs. Making stuff that takes a lot of effort virtually effortless is a massive gain to humanity, especially for people who depend on technology like this to access media (especially so for media in the public domain).
It's such a weird argument that honestly it feels like a trap designed to hurt the striking actors. I hope none them are stupid enough to latch onto this article to support their argument because as an actor saying you're afraid a robot might be better at understanding and conveying the emotion in a book is not a good look. Even more so when you're aiming at accessibility software.
You'll see this happen whenever the tech becomes slightly widespread. When DALL-E was the only kid on the block, and stable diffusion and midjourney and such barely existed, the media was breathless with excitement over how cool these tools were. Then as soon as sd came out, and made image generation available to a ton more people, without corporate censorship and control, you started to hear about how bad it was, how it was undermining artists, and so forth
>it almost seems to be trying to hurt the cause of the actors currently on strike.
Thats a lot less strange then it sounds. Sabotage is a real strategy and it has become incredibly easy with increasing tribalism and the inability to call peoples stupidity out. The rational goes the more i can make your side look and act stupid, the better chances for my side. It becomes increasingly easy if there arent any checks and balances in your ingroup. It of course leaves everyone in a worse situation but a lot of people dont seem to care.
This gets compounded because there is a blind spot to bad behavior of ingroup people. You know plenty of people on your side who aren’t crazy, so you dismiss the extremists as crazy people on the periphery, while those outside the group see them as examples of the norm of your group.
Yes. Unfortunately none of this is easy. Even if you are sure you are not in a deteriorating cult (pleasant one or not) there is still a ton of blind spots not stemming from social environment but your ego for the lack of a better term. Cognitive biases are a real danger and both advertisement and cognitive warfare target those.
This is very much an asymmetric struggle. While you might be a sensible person when you are paying attention to the right aspects, most of the time you arent. Or to quote the Provisional IRA of all people
>Today we were unlucky, but remember we only have to be lucky once. You will have to be lucky always.
I dont see a way around regularly checking if you have gotten unlucky and carried away. Its really easy to just be opposed to a caricature, figuring out and accepting that you have become one as well as a reaction is quite a lot harder. Mitchell and Webb made a great bit about checking regularly if you and your friends are maybe the baddies.
https://www.youtube.com/watch?v=ToKcmnrE5oY
If people around you take offense to you doing that, thats a very valuable indicator. If this leaves you with a collapsed reality model, dont worry, thats normal. Stuff is just complicated. Fortunately less bad is often a very reachable bar.
And doubly so for older, long-tail content which would otherwise be unlikely to get an audiobook produced of it. There's more to public-domain English fiction than Shakespeare and Dickens, but no one's stepping up to voice-act audiobooks for obscure authors from the 1800s.
Well, I have seen some translators blame translation models for stealing their jobs, especially nowadays that these services like Google Translate, Deepl are based on AI, people are creating parallelism with the ongoing controversy around AI art etc... And attract a lot of attention on Twitter.
But since I'm not a native English speaker and these models were incredible for learning the language, also speech-to-text models, all these accessibility models are used in the vast majority of cases where there would be no human alternative, for example transcribing all Youtube videos, or correcting the grammar of this HN comment, so I really agree with you.
"Making stuff that takes a lot of effort virtually effortless is a massive gain to humanity"
Sure, and having AI generated movies and stuff would reduce the physical and human resources needed to make movies. In either case, the people who lose those jobs won't be happy about it, despite the gain to humanity.
It should not be overlooked that this is less of a problem of automation but one of how we organize society. What do we do once large parts of society become redundant? With many of them already having had bullshit jobs doing busywork? The effect of substantial numbers of high paying jobs vanishing alone will be disastrous for the economy. Technological progress is just highly deflationary.
Well they don't have to be happy about it, heck I think it's more than fair they protest when technology impacts their livelihoods. There's even some valid concerns about the way some of these models are trained or exploited.
But this instance is a terrible fight to pick. You'll be pitting a fight against a non-profit, accessibility software, a public domain library, on the basis that a computer is doing too good a job reading a book out loud.
It’s a non-profit, open source venture with negligible revenue doing the best it can to open up reading to people of all colors and stripes. What a stupid fight to pick.
If you want to complain, complain about the Bezos-owned WaPo adding machine-generated audio versions of all their stories (linked prominently under the headline) and pushing them for a podcast-like experience (terrible quality generated audio, might I add, having clicked on it yesterday by accident while scrolling and couldn’t figure out how to stop it until I just killed the app).
Unlike Project Gutenberg, WaPo and NYT used to hire narrators for their popular stories or bleeding headlines. That is revenue and jobs that might actually be erased.
Missing from this article: any quotes from striking actors or union organizers actually saying that they oppose Project Gutenberg's efforts specifically.
AI is one part of why actors are striking, and their fears are pretty directed at studios and employers. In particular they're worried about contract clauses that grant studios perpetual rights to use their likenesses and voices to train AIs that studios hope will replace them. It is a convenient narrative to phrase that as if they are opposed to AI entirely or that they are opposed to accessibility or voice-over technology. In reality even if actors were fully embracing AI-generated content, the terms being proposed by studios for ownership over training data and unlimited monopolization of actor voices and performances would still be egregious.
There have been conflicts about text-to-speech efforts in ebooks before, but it's not clear to me that the majority of actors have any problem with stilted, lifeless text-to-speech designed by a nonprofit to increase ebook accessibility.
And maybe actors will be furious at Gutenberg, I don't know. But that's the point, a good way of figuring out whether or not striking actors are mad at Gutenberg would be to talk to them directly, which this article doesn't do. What I'm saying is, in the middle of a highly public strike, be careful of uncritically accepting the word of writers who are not part of that strike when they tell you what the strike is about and what the people striking believe and want.
You're making the same mistake the article is making.
There is a huge effort going on to frame striking actors as if they're anti-technology and to turn what should be a conversation about clearly unacceptable studio terms around IP rights into a philosophical debate about the nature of progress. If the union organizers all come out and start condemning Project Gutenberg then fine, we can have a conversation about Gutenberg then. But as far as I can tell online, that hasn't happened yet.
And even if we do end up having that conversation, it's not at all clear that striking actors are going to feel the same way about nonprofit accessibility tools as they would feel about purely profit-motivated projects for Amazon books or (more relevant to the actual strike going on) movie studios. These conversations oversimplify what actor demands actually are.
Before you get into a giant argument, check to make sure that the argument is actually happening and isn't just the result of a single Internet article by a single writer. Made-up conflicts online started with "people on Twitter say" junk pieces that would quote single-digit follower Twitter accounts as if they represented a consensus, and have now moved past that into pieces just kind of confidently stating what demographics believe with literally no sources at all. If union organizers have a problem with Project Gutenberg, they will say something about it at some point.
And maybe they will! But I can't find a source on that yet; as far as I can tell from some quick searches we have one isolated article online framing this as a major conflict with unions even though there are zero actual union organizers in the article saying that it's a conflict.
Why are people (or maybe it's AI, ha!) writing hit pieces about a non-profit who attempts to make classical works widely and freely available? Not sure who/what "quartz" is but it's giving off major writing-for-hire vibes. Likely there is not a single member of SAG who cares what Project Gutenberg does.
Maybe this is another union ploy, similar to the Netflix hit pieces a month ago about the ML job.
I just listened to the beginning of one of these books.
The book text starts out "I." as in, roman numeral for chapter 1. The audio book read this as "eye" (or "I", but not 1).
Then the third sentence in the text says "1793-05". The audio was "1793 to 2005". Actually it was 1793 to 1805.
I listened to a few more sentences and had an idea of the state of it currently. The tech has advanced in the past few years, still has a way to go. Insofar as striking actors, they probably should be worried about what the big companies are up to.
> Then the third sentence in the text says "1793-05". The audio was "1793 to 2005". Actually it was 1793 to 1805.
Bit of a tangent here but did you get 1805 from context or is YYYY-YY a common year range notation that I'm not aware of? Because I would never guess 1793-05 meant 1793-1805 either.
1898-02 or 1998-02 would be the years someone attended high school or college. The context of when it was written, either in 1905 or 2023 can be important. Seems to be pretty common notation when spanning centuries, particularly +/- 10 years. I grew up during the turn of the century and seemed to be quite common.
It’s not but it’s obvious from the context, which I think is the point. A GPT-4 LLM could probably figure it out from context, but most text to speech models probably don’t have as much semantic training.
Wouldn't that be caused by an OCR issue? No matter how primitive I doubt a TTS model would mistake I for 1. The book was probably poorly scanned, OCRed, and then had automated audiogen.
> No matter how primitive I doubt a TTS model would mistake I for 1.
"I" is the correct text, but it's intended as a Roman numeral and should be read as "one". (And, a few chapters later, "IV" should be read as "four", not "intravenous".)
Listening to an audiobook narrated by an AI voice can get old after a few hours. There is absolutely still hope for traditional voice narration, especially if different voice actors want to voice different characters in the story.
I imagine this will sit where the LibriVox recordings did before. "Good enough" to cover the vast amount of works that never received studio treatment, but not anything that will ever replace a human at the high end. Even if the likeness is absolutely perfect, people will still prefer to know they are actually listening to their favorite actor.
People can listen to every book they want in decent quality instead having to wait for it to be adapted as an audio book. Of course blind people would especially welcome this. Audio books would get cheaper, which seems like a desirable trait to me. In the end, all of these AI "conversions" are menial in a way anyway, so why not automate it. Having access to these automations, hopefully subscription free, would also be a kind of freedom for everyone. The freedom to consume information in any form you want. A universal translator isn't all that different from this in that regard.
As for actual creative tasks, that is another matter altogether. I can only hope that real creativity doesn't die out and that TV series and other things don't get any more homogenous than they already have become. At least there are still enjoyable outliers in this plethora of entertainment output. And in some way, audio books can also be creative if they really add emotions that aren't written down exactly. Similar to how the same sheet music can be adapted by different musicians in slightly different ways or instruments and such. Even if all of that could be synthesized, what it means is that selection of the results would become the creative task. Just like stable diffusion and others output hundreds of images for a prompt but only a few are actually interesting. By definition of this automation, you will always end up with more than you can consume and finding the right result for the right people then becomes the next hurdle.
I still wonder though, whether new entertainment really is necessary until the end of times. There probably are already more books and music in the world to last you a lifetime. And after 100 to 200 years of continuous TV and movie output, you probably could just reshow the ones from a decade ago for the next generation, no need to rehash everything. This assumes that there are no new media though, such as maybe walkable VR movies. But even film from decades ago can be remastered in 4k and I begin to doubt that anything above 4k makes much sense. Even 2k is still perfectly fine to me.
We have millions of digitized books by now. Only a very very small percentage of those are worth spending the money to turn into an audiobook narrated by a human.
A lot of people do not read books (for various reasons: they are blind, they are driving, they are jogging, they are doing chores, they just don't enjoy reading), leaving millions of pieces of art inaccessible to them.
Getting TTS quality up to near human narration level means absolutely everything will become available to them overnight.
To eliminate work of course. To advance the state of technology where everyone can have the base essentials for free, or as low a cost as physically possible.
You should listen to these Project Gutenberg audiotexts…they are really pretty good. I would have no problem listening to them for an extended period. The real challenge Project Gutenberg faced was automating the removal of the random metadata noise from the original plain text versions of the content. And it will continually improve from acceptably good to excellent in short order with perhaps a tiny amount of hand tuning.
It isn’t just the blind who benefit. As you age, eyesight degrades and the mind dulls. At some point, audiobooks are a real boon, especially with PD texts where you can read until you get tired, then listen. Also the growing demographic of live-alone elderly like the background sounds of people talking in their residence, explaining part of the popularity of talk radio. They feel a little less lonely. These audiotexts provide an intellectually varied and overall healthier option. It seems like an Absolute Good to me.
AI audiobooks also seem they’re ripe for “perfection over time” where parts are indicated where it’s not pronounced right or needs additional inflection, etc, and that can be rerendered.
There will still be a market for premium audiobooks with an engaging narrator.
This will help fill the gap for books that would never have an audiobook, or for people that never would have paid anyway (so they aren’t lost customers).
edit: I'm not sure what's particularly exciting about this paper otherwise, it just seems like they parsed the PG html format and then applied some previously developed speech model to them.
By the end of the decade, end of next decade tops, text to speech will be a solved problem, anyone who thinks otherwise hasn’t been paying attention.
It suck for voice actors. If they’re lucky the top percent of them might retain a job due to the novelty or branding of their voice, but most will be unemployed.
edit: erroneously said dictation (which also will be solved :p)
I listened to Call of the Wild and it was excellent. I think some of the intonation and contextual emphasis, tone etc could be improved. I suspect this is definitely the nail in the coffin for artisanal narration. Some of the recent SOTA models (below) do very well in these areas, so I’ll wager in 5 years all ebooks will be narratable via a local or cloud model as part of the reader.
I was under the impression that striking actors concerns with AI were over AI reproducing their likenesses, but this article is claiming that the fears would also consider non-specific (non-imitation, non-likeness, ?) generated voices a threat. The former is at least what I've read as far as union demands go.
Actors demanding a cut of revenue generated with their likeness I can understand, but trying to stop all AI-gen content, even that which does not imitate a real person and was trained on ethically sourced, non-SAG data, smells like business-model protectionism or even luddism.
The article is just wrong. SAG is trying to prevent the specific companies that hire their members from recreating a performers actual visual or vocal likeness without compensation or for very little compensation. To the extent they oppose a “generic” AI voice it would be in the context of that voice being trained on material that was originaly created by SAG performers or under SAG contracts and used for training with little or no compensation.
> What the reading does lack is the versatility and depth that human actors bring to the same work. It’s a far cry from, say, Stephen Fry, who created a unique voice for dozens of characters while reading the Harry Potter audiobooks. But Project Gutenberg’s audiobooks are free, the process is quick, and the text-to-speech system used is scalable.
You loved cheap tasteless hydroponic tomatoes? You'll love cheap bland voice-acting.
You might not be good at reading, blind, do not have the time to read and the money to buy audiobooks. It's far from perfect, but it's better than what these people had before.
Making these works available to most people is the goal of Project Gutenberg and this is another step in that direction.
They do not have the money to hire voice actors to record audiobooks even for their most popular titles
You're missing the forest for the tree here: my point isn't about this particular case or project Gutenberg, it's just that this example shows us the future, this is a future that looks very much like everything automation touched before, where we end-up with mass-produced cheap crap instead of simply better things at the same price or more affordable stuff with the same quality.
In a competitive market, bad product drives out good.
I can support Stephen Fry's right not to have an Ai version of his voice created without his permission.
> Speaking at the tech-based CogX festival in London, he played a clip of a history documentary that faked his voice, without his knowledge. It appears as though the actor is narrating the show but the AI-generated voice was actually created by technology learning from Fry reading all seven of the Harry Potter audiobooks.
However, the notion that it's somehow nefarious to create a new voice to read audiobooks (with the permission of the person supplying the training data) is ridiculous.
This is called progress. Maybe not today or tomorrow, but in a few years the trend is to have all readers replaced by AI once technology will make their voice credible, then all singers will face the same fate. I'm sorry, but in a few decades, possibly years, all movies and music will contain more and more AI generated content with fake actors and artists who never existed whose fame has been carefully planned on social media.
Few will complain: "so what? We can still act or make music for fun"; then one day we'll see virtual politicians controlled by the same small oligarchy of very wealthy people, with the difference that a virtual politician could be taken out and swapped to another one constructed to appeal the right target just by filling a prompt.
Sadly, a big percentage of recent movies or music hits is already a mashup of old original ideas, automating the "creation" process to increase profits is just going one step further. Indie or niche movies or bands won't be touched simply because that's not where the money is.
Ok, show me the actor who is afraid of losing a meal ticket by not getting to bring their talent to an audiobook reading of Discourses on the First Decade of Titus Livius.
Do the author and the imaginary picketline behind them take issue with Librivox recordings of public domain works?
This ship has sailed and worrying about AI copying a particular actor or voice of a professional actor is virtually meaningless long term.
AI will enable the creation of voices and people who never were and even the corpus of people used to train the AI wont need to include actors.
If you read books out loud or are an extra you don't and honestly shouldn't expect an income from it and this problem stands to get worse not better.
These is little political will to implement UBI so best get to learning a new profession whilst audio books transition to tools produce dozens of professional virtual voice actors for 19.99 or free if you dont mind user unfriendly open source you have to set up yourself.
> If you read books out loud or are an extra you don't and honestly shouldn't expect an income from it
It's the other way around. If people cannot expect an income from doing work, the work will be left undone. Or it will be done poorly by cut rate profiteers.
There wont be much space for people doing this for work for money in 10 years. There is no reason to believe the quality of automation will continue to be poor any more than we would have been reasonable to believe that chess engines would continue to perform poorly.
I hadn't noticed chess engines had replaced human players. Do we have more or less professional human players today than back when Kasparov lost that historic game?
The point is that primitive engines were replaced by more advanced engines in time that surpassed humans first running on super computers and now on laptops.
Computers haven't replaced chess players because chess is recreation it is an end unto itself not a means to an end.
With voice acting it is the means toward an end which is the finished audio file which can be licenced to consumers for money.
The means costs about 300 usd per hour of finished audio whereas there is every reason to believe software can drive the per unit cost so low its impossible for a human to make a living whilst still being profitable.
Eg most of the work could end up open source leaving someone to provide 2 people to burn 6 months to produce millions of hours of audio. Lets value their time ar 150k per year and suppose they produce 3 million hours the labor cost is roughly 5c per hour.
They can probably afford to offer 4.99 per hour while being wildly profitable even after other costs are considered.
A human being doing a dozen hours a week wouldn't be able to afford a good enough tent to impress the other hobos.
I get that you think software can only improve because it evidently gets more efficient and fast over time and that means anything humans do for money is liable to be replaced by computers. What I think you fail to realize is that better technology enables people to become better at what they want to do.
There are now more _professional_ chess players than in the 90s. Because people like to play chess? That's circular reasoning isn't it? People can train and learn from software, with the end goal of being better players and the economics of the game have because of this absorbed more people.
When you take a snapshot of a field of human activity and imagine the next one, you guess there will be less people and more machines. This is naive and unfounded in history. There are now more fashion designers and tailors and a more complex sartorial economy than before robotic looms. Even accounting for population growth. There won't be less voice actors in a generation because of robotic voice acting. There will be a more diverse and more complex economy around producing voice content.
Predicting machines will replace people is predicated on a industrial view of the economy where people are either capitalists or workers and capitalists vie to use capital to reduce labor costs. In practice, many trade-offs apply, and often the introduction of technology changes dynamics in unpredictable ways. Not everything in the economy is some form of inefficient manual labor that can be profitably replaced by a box.
This whole "entitled workers shouldn't expect to deserve an income" smacks of elitist propaganda. Technology changes culture and economy in complex and nuanced ways, and the caricature of workers mad at clever engineers robbing them of cushy jobs is usually just a front for selling some mass produced lower grade substitute. And the people working on the high quality stuff don't get replaced by robots. They learn how to use the technology to innovate and add more value, out-competing these sorts.
Take the auto industry for example. They went heavily into robotics for decades, automated welding heavily and raced to the bottom culminating in a heavy series of recalls in early 00s. Meanwhile, skilled welders have become some of the highest payed industrial workers and what did the auto industry do to climb up the value ladder again? Crammed their cars full of non-robot made software.
I believe the human voice-actors are justified within this context. No comment on anything else but how does a “free market” come into play with this? It seems consumers should be able to choose between an AI narrated and human narrated audio book - given equal access to both.
I personally am not a big fan of AI but playing devil’s advocate to a degree it seems AI narrated books may be useful for translating works into less common languages.
I hope we can widely implement “generated by AI” labeling for works/items to notify consumers.
Edit: fwiw I have a difficult time listening to audio books read by humans in the first place. But that’s largely irrelevant
> It seems consumers should be able to choose between an AI narrated and human narrated audio book - given equal access to both.
If humans want human narrated books, they'll get access to both. Human narration will only disappear if people aren't willing to pay for it. I'm pretty sure that ultimately AI narration will be so much better that human reading will be a niche thing, but I view that as a positive overall. More books get read with high quality narration, and consumers don't have to cover the cost of a human narrator.
Obviously unfortunate for the human narrators, but that's why we should be pushing for UBI - AI and robots are going to eliminate a lot of professions, and I think it's wrong to keep jobs around for no other reason than to keep people employed, especially when doing so ultimately leads consumer costs to be higher. You see this in lots of places now - a big one is the teamsters using their union power to keep automation out of our ports. Literally everyone in the US could benefit from goods moving more quickly, efficiently and cheaply through our transportation infrastructure, but instead everyone has to pay a bit more so humans can keep unnecessary jobs.
I've found a few audiobooks on Audible with narration so bad (mostly failure to properly parse sentences, putting emphasis in the wrong place) that an AI version made by random people on YouTube is superior.
For non-bestsellers, I think it'll be 100% AI narration. If human narration remains, it'll be top-tier books read by the most famous/best voice actors/narrators.
Free audiobooks narrated by amateurs (who are real humans) already exist. Just go to YouTube and search for classic book name + "audiobook". Yet people go to Audible for those books because there is a difference between amateur work and professional work. Same as why people buy Shakespeare books with annotations instead of just using some random online version.
Spreading FUD about text-to-speech technology that is not completely lifeless is a bit like blaming google translate or OCR programs from taking away jobs. Making stuff that takes a lot of effort virtually effortless is a massive gain to humanity, especially for people who depend on technology like this to access media (especially so for media in the public domain).
It's such a weird argument that honestly it feels like a trap designed to hurt the striking actors. I hope none them are stupid enough to latch onto this article to support their argument because as an actor saying you're afraid a robot might be better at understanding and conveying the emotion in a book is not a good look. Even more so when you're aiming at accessibility software.