Recording, editing and finding new ways to bring written stories to life.

There is a lot that goes into making an audiobook. Someone (or something) has to bring the words to life, the recording needs to sound good, and the finished chapters have to be put together so that people can listen without interruptions or distractions. Even a relatively straightforward project can involve hours of recording, editing and preparation.

This fall and winter, our program is working on ways to make that process more manageable. We’re recording human narration, experimenting with AI-assisted audio and developing tools to help turn written material into finished recordings. We’re interested in the creative side of the work, but also in the practical details that make a difference once an audiobook leaves the studio.

That means dealing with sound levels, file formats, background noise, chapter breaks and publishing requirements. It also means figuring out which tasks can be made easier with technology and which still need a person to listen, make decisions and get things right.

Getting the Sound Right

A voice recording can sound fine when you first play it back and still need quite a bit of work before it is ready to publish. The volume might vary between sections, a pause might feel awkward, or a faint background noise might become distracting over a longer recording.

The technical settings matter, too. Audiobook distributors specify requirements for the files they accept, including sample rate, encoding format and audio levels. ACX, for example, requires submitted audio to use a 44.1 kHz sample rate. When a recording starts at a different rate, it needs to be converted appropriately before delivery.

Changing the sample rate doesn’t magically improve the original recording. What matters is doing the conversion properly, without introducing unnecessary distortion or other problems.

File size is another consideration. MP3 remains a common format for audiobook distribution, and a constant bitrate of 192 kbps is one setting we’re using in our production workflow. It offers a practical balance between file size and sound quality for spoken-word material. The best settings still depend on the source recording and the requirements of the intended platform.

These details may seem minor when you’re working with a short passage. When you’re putting together hours of narration, consistency becomes much more important. A reliable process helps prevent small technical problems from multiplying across an entire book.

Getting the Volume Right

One of the trickier parts of audiobook production is finding the right volume. A recording needs to be comfortable to listen to, but turning everything up isn’t the answer. Push the sound too hard and the voice can lose its natural dynamics. Leave it too quiet and listeners may struggle to hear it in ordinary surroundings.

ACX specifies an RMS level between −23 and −18 dBFS. RMS is a way of measuring the average energy of an audio signal, giving us a useful indication of its overall level. Our workflow uses −20 dBFS as a target when normalizing the audio, placing it near the middle of that range.

Peak levels need attention as well. A recording might have an acceptable average volume but still contain a sudden spike that causes distortion. ACX requires peaks to remain below −3 dBFS, so our workflow aims for a more conservative ceiling of −4 dBFS when applying a limiter.

A limiter controls peaks that rise above a chosen level. Used carefully, it can help prevent clipping while leaving the voice sounding natural. Used too aggressively, it can flatten the performance and make listening tiring.

There is another wrinkle: MP3 encoding can sometimes produce reconstructed peaks that are higher than those measured in the original audio. Leaving some headroom helps reduce that risk, but it isn’t a guarantee. The finished MP3 still needs to be checked after encoding.

We’re working to make these measurements and adjustments part of a repeatable process. The numbers give us a starting point; listening to the result tells us whether the recording actually sounds right.

The Importance of Pauses and Background Noise

Good narration needs room to breathe. A pause can give a sentence weight, allow an idea to settle or signal that a chapter has ended. Cut those moments too short and the recording can feel rushed. Leave too much silence and listeners may wonder whether playback has stopped.

Our production workflow experiments with adding consistent pauses at the beginning and end of audio files. One configuration uses 0.8 seconds at the start and three seconds at the end. These timings give us a practical starting point when preparing chapters, although the best pacing depends on the material and the requirements of the distributor.

We also pay attention to the noise floor: the level of background sound beneath the narration. ACX requires the noise floor to be no higher than −60 dBFS. That is a useful benchmark when checking for persistent hiss, hum or other unwanted sounds that can become noticeable during quieter passages.

Clean digital silence helps keep chapter openings and endings quiet, but it doesn’t remove noise from the recording itself. If a microphone has picked up a fan, traffic or room noise, those sounds need to be addressed separately. Noise reduction also needs care, since too much processing can leave a voice sounding thin or unnatural.

These are the little things that become obvious over several hours of listening. Consistent levels, sensible pauses and a quiet background help people settle into the story rather than notice the machinery behind it.

Human Voices and AI-Assisted Narration

Human narration remains an important part of our work. A narrator brings their own voice, timing and interpretation to a piece of writing. The way someone delivers a line can change its meaning, give a character personality or make a passage feel intimate and immediate.

At the same time, some projects involve a great deal of written material. Recording every page by hand can require substantial time and resources, particularly when the material runs to hundreds of hours. AI-assisted narration gives us another way to approach some of that work.

We’re interested in where the different methods make sense. A story that depends on performance may call for a human narrator. A large collection of reference material or selected archival documents might be suited to a different approach. Some projects may benefit from a combination.

AI-generated speech still needs to be checked. Pronunciation can be wrong, pauses can land in strange places, and a passage that sounds convincing on its own may become monotonous over a long listening session. Generating the audio is only one part of the job.

The point is not to replace human narration. It is to have more options when deciding how to produce a recording, particularly when the amount of material makes a conventional approach difficult to manage.

Turning Recordings into Finished Audiobooks

Once the audio is ready, there is still the question of how to package it.

An audiobook needs more than a folder full of sound files. The chapters must be arranged in the right order, the publication needs identifying information, and listeners need a straightforward way to navigate the material.

This is where digital publishing standards come in. The World Wide Web Consortium (W3C) has developed recommendations for digital audiobooks, including a Lightweight Packaging Format (LPF) for organizing publication files and their associated information.

Depending on the specification, a package can include audio tracks, a publication manifest, reading-order information, navigation and cover artwork. These components help compatible software understand what the publication contains and how its parts fit together.

Getting the details right matters. Putting a collection of files into a ZIP archive doesn’t automatically make it a compliant audiobook package. The required structure, metadata and file references need to follow the applicable specification.

We’re looking at ways to simplify this part of production, including workflows for recordings that already exist. Someone who has spent weeks recording a book shouldn’t have to start again just to organize the chapters and prepare the material for digital distribution.

This can be particularly useful for independent writers, artists and community organizations that don’t have a dedicated production team.

Making More of the Work We Already Have

Not every audio project starts with a new recording. There may already be interviews, readings, performances, scripts or community records stored on a computer or tucked away in an archive.

Sometimes the work begins with sorting through those files and deciding what is worth preserving. Other times, recordings need to be divided into chapters, given proper titles or converted into a format that can be shared more easily.

Written material presents a similar opportunity. Stories, educational resources and local histories can remain on a shelf or hard drive for years. Making an audio version may give people another way to engage with them, whether they’re listening at home, travelling or taking a break from a screen.

Not every document needs to become an audiobook, of course. Some material is best left as text, and some deserves the care of a human performance. The value lies in having more choices about how something can be presented.

For community archives and independent creative projects, those choices can matter. Useful knowledge doesn’t always arrive in a neat, publication-ready package. Sometimes the first step is simply finding a practical way to work with what is already there.

Building Our Own Tools Along the Way

Part of this work involves developing and testing our own digital tools rather than relying entirely on ready-made services.

That means getting into the details: trying different audio settings, organizing files, checking exports and finding ways to automate repetitive tasks. It also means figuring out where automation isn’t enough and someone needs to listen carefully or make a judgement call.

Some parts of the process can be handled locally in a web browser, including certain kinds of audio organization and packaging. Where a workflow runs entirely on the user’s device, it can avoid sending those files to a remote service. The precise capabilities depend on how the tool is built, but local processing is worth considering when working with personal, creative or archival material.

There is a practical side to building these tools, too. Creative work involves managing costs, time and limited resources. A workflow that saves unnecessary effort can leave more time for recording, writing, editing and working with people.

For our program, this is part of learning by doing. We can start with a problem, try an approach, see what breaks and make improvements. Sometimes the most useful result is a working tool. Sometimes it is a better understanding of what a project actually requires.

Either way, the experience builds skills we can carry into other work.

Where We Go From Here

Our work will continue through the fall and winter as we record, experiment with narration and improve the way we prepare audio for publication.

Some projects will call for human voices and the care that goes into a performance. Others may benefit from AI-assisted narration, particularly when there is a large amount of text to work through. Whatever the approach, the finished recording still needs to be checked, edited and prepared for the people who will listen.

We’re still working through the details, and not every experiment will produce the result we expected. That’s normal when building something new. Each attempt gives us a better sense of what works, what needs attention and where we should put our effort next.