AI Audiobooks: What Human Narrators Still Bring to the Story

AI narration is developing quickly, but the audiobook industry is about much more than turning written words into speech.

In Episode 5 of HIGH NOTES, Melissa Thom speaks with Dan Jones, former Head of Audio Services at Hachette UK, about how audiobooks are actually made, where AI is entering the production pipeline, and what that means for human narrators.

Dan spent more than two decades at Hachette UK, including seven years heading its audio services operation. His role included overseeing casting and production teams, developing the technical and creative strategy for the studios, and managing a network of recording facilities in the UK and internationally.

That experience gives him a useful perspective on a question that can easily become oversimplified: if AI can now produce an entire audiobook, does that mean it can replace what a narrator actually does?

Casting an audiobook is not simply choosing a good voice

From the outside, audiobook casting can look like a purely creative decision. Dan’s experience at a major publisher paints a more complicated picture.

Budget comes first. Many titles fall into established per-finished-hour rates, with publishers and production teams already familiar with narrators, agents, and the rates associated with different levels of experience and demand. Other books sit higher up the budget scale. A television or film tie-in might lead the publisher towards an actor associated with that production, while major bestsellers may have enough budget to attract significantly bigger names. Throughout this, the publisher and author remain closely involved, with the author usually having final sign-off on the chosen narrator.

But the process is not purely transactional.

Dan describes casting as a process of discovery and relationship-building. Producers remember which performers suit certain material, build relationships with agents, and gradually develop a sense of who might be right for future projects.

That kind of judgement matters when considering AI narration. Producing speech is only one part of creating an audiobook. Someone still has to decide what kind of voice belongs with the story in the first place.

Is AI narration a performance, or a rendering?

Melissa asks Dan about an AI replica of Michael Caine narrating The Odyssey. His response goes directly to the heart of the human-versus-AI question.

“Is it a performance?” he asks.

For Dan, it is closer to a rendering.

He points to the lack of dynamic range, the absence of genuinely quiet or loud moments, lines that don't quite land, and intonation that still feels different from human speech. Music and sound design can improve the experience, but he suspects they may also help disguise some of the weaknesses in the synthetic narration.

In his view, AI struggles most with the hardest-to-quantify part: “the human-y bit, the heart, the soul”.

Melissa heard something similar when she listened. The cadence repeated, the speed made the ideas harder to absorb, and the performance didn't seem to breathe the way a human reading does.

Dan acknowledges that newer AI models are significantly better than those available only a few years ago. Pauses and elements of natural cadence have improved. For some listeners, particularly when an audiobook is playing in the background, the result may already be good enough.

But “good enough” and “performance” are not necessarily the same thing.

Humans currently have the quality advantage. AI has the cost advantage

The obvious question is whether certain types of books suit AI narration better than others.

Backlist titles, textbooks and academic material are often suggested because AI can dramatically reduce the cost of producing them. Dan understands the argument, but he is not convinced it is that simple.

Educational material still needs to engage the listener. A low-cost voice does not necessarily help if the result makes the material harder to follow.

He recalls discussions about using AI to test older titles, potentially recording them later with human narrators if sales justified the additional investment. Voice cloning may also offer options when a particular performer can't, or doesn't want to, spend days recording in a studio.

For now, though, Dan sums up the balance simply: human narrators still win on quality and performance, while AI wins on cost. The gap is narrowing, but the two strengths are still different.

Where AI narration is actually appearing

Dan makes an important distinction between the traditional trade publishing market and other parts of the audiobook ecosystem.

During his time at Hachette UK, AI narration was primarily an area of experimentation and research rather than commercial production. He says that, in his experience, much of the current activity sits outside mainstream trade publishing, particularly within self-publishing and technology-led platforms.

That distinction matters because publishing involves more than generating audio.

Editors, authors, marketers, publicists and producers all contribute information and judgement about a book. Those decisions influence casting, tone, positioning and ultimately how the story reaches an audience.

Dan suggests that technology companies may be making more binary decisions: choose a voice and generate the book. A traditional publisher would be more likely to ask whether that voice was actually right for the material.

Even if AI does the narration, a creative question remains: who, or what, should be cast.

The most useful AI may not be behind the microphone

One of the clearest distinctions in the conversation is between using AI to replace a narrator and using it elsewhere in the production process. Dan is much more enthusiastic about the latter.

Automated tools can reduce production costs, speed up editing, identify errors, and perform an initial quality-control pass before a human checks the work. Machine learning has also transformed noise reduction and audio repair, making it possible to improve recordings that would previously have required far more technical intervention.

For Dan, this is an area where technology can genuinely improve audiobook production while leaving the creative relationship between narrator and producer intact.

The narrator still performs. The producer still shapes the work. AI takes on some of the repetitive or technical labour surrounding them. This model differs markedly from replacing the human voice altogether.

It also connects with something Melissa emphasises in BRAVA’s training: skilled performers can themselves reduce production time.

Understanding microphone technique, breathing, preparation, and the specific demands of audio acting makes a narrator more efficient in the studio. Technology can improve the workflow, but so can craft.

New technology does not always replace the old format

Dan compares the arrival of AI audio with the earlier growth of ebooks.

When the Kindle appeared, people predicted printed books would disappear. That did not happen. Ebooks became another format, serving different customers and different circumstances.

AI audio may develop similarly.

Dan can imagine synthetic narration growing within particular markets and ecosystems without necessarily replacing traditional audiobooks. Audiobook listening itself has expanded as technology such as smartphones, Bluetooth and smart speakers has made it easier to fit audio into everyday life.

Rather than one format automatically eliminating another, Dan suggests human and AI narration may coexist.

The question may ultimately become less about which technology “wins” and more about which version a particular listener wants.

Audiobook narration is a craft, not simply reading aloud

The episode also explores what publishers actually need from human narrators.

At Hachette UK, Dan’s studios frequently recorded authors rather than professional performers. Some years, he estimates, authors accounted for around 60 per cent of the books produced.

That meant production teams worked with people who might have little or no prior microphone experience. It also created opportunities to broaden the range of voices represented in audiobooks and cast people who were right for the material even if they were less experienced.

But reading well for a few minutes is very different from sustaining a performance throughout an audiobook.

Dan describes a simple test used in casting workshops: someone may sound good for five minutes, but what happens after half an hour?

A professional recording day can mean hours behind the microphone. Narrators need stamina, preparation, and the ability to maintain the performance without exhausting their voice or losing connection with the text.

Melissa adds that performers need to understand breath, pickups, preparation, and the pressures of working while the studio clock is running.

These are learnable skills. But they are skills nonetheless.

What happens in the quiet moments?

The differences become especially apparent in quieter passages.

Melissa asks Dan what a great narrator is doing in those moments that a machine cannot.

His answer begins with being natural and understanding the text, but quickly becomes something broader.

A human performer draws on experience. They think about what a character is feeling. They respond to the words through everything that has happened to them before they entered the studio.

Those experiences affect decisions that may be almost impossible to isolate: where to breathe, where to hesitate, which word to lean on, when to slow down and when not to.

Dan compares it to screen acting. A screenplay does not need to describe every emotion because an actor can communicate meaning through a glance or a movement.

Audiobook performance has its own equivalent, expressed through the voice.

AI can potentially clone the sound of somebody’s voice. Dan’s point is that it cannot simply clone the life that shaped how that person uses it.

“Even if they cloned all your voices,” he says, “they’re not going to be able to clone all your lived experience.”

That may be one of the clearest explanations of what human narration still contributes.

The word is connection

When Melissa asks Dan to reduce that contribution to one word, they arrive at “connection”.

Books let us explore an inner life that everyday conversation doesn't always make accessible. For Melissa, knowing that another human being has also encountered and interpreted those words is an important part of the experience.

That relationship exists between author, narrator, and listener.

Dan uses the Beastie Boys Book as an example of an audiobook whose appeal comes partly from the people involved and the relationships between them. The personalities and shared history are part of the storytelling itself.

That kind of production illustrates why evaluating an audiobook purely by whether the words were pronounced correctly misses so much of what audio storytelling can be.

What does the future look like for human narrators?

Despite the rapid improvement in AI narration, Dan does not describe a collapse in demand for human narrators. From his perspective, work in traditional publishing has remained relatively stable and may even have increased. His hope for the future is straightforward: he wants real people, with real lived experience, continuing to come through the microphone.

But he would also like the industry to become more ambitious.

Rather than simply producing more conventional audiobooks, Dan sees potential for new forms of audio storytelling, oral histories and long-form productions that take advantage of increasingly accessible technology.

He draws a parallel with music production, where the ability to create professional work outside traditional studios helped open up entirely new possibilities.

Perhaps the most interesting future for audiobook technology, then, is not one in which a machine perfectly recreates the traditional narrator.

It may be one in which cheaper and more accessible production tools allow humans to make forms of audio storytelling that previously would not have been possible. AI can help with parts of that process. It may even become another choice for listeners.

But Dan’s argument throughout the conversation is that narration at its best remains an act of interpretation. The words matter, but so does the person encountering them.

And that is the part of the audiobook pipeline that is much harder to automate.

Listen to Episode 5 of HIGH NOTES, The Audiobook Pipeline, with Dan Jones for the full conversation about AI narration, audiobook production, and the future of human performance.