Guides
AI narrator: how to make AI narration sound like a real narrator
The Voizum team · Updated · 9 min read
An AI narrator reads your script and hands you an MP3. Whether it sounds like a narrator or a screen reader depends on the voice you pick, how you write the script and the pace you set. Below: what a documentary, a bedtime story, an audiobook or an explainer video needs, how to mark pauses in the text, and how to do it in Voizum.
What an AI narrator is, and what makes AI narration sound professional
Paste a script, pick a voice, download the audio. That's an AI narration generator. Current voices breathe, lift at questions and stop at periods, so the robotic screen reader of a few years ago isn't the benchmark anymore.
What separates a machine voice from a narrator comes down to three calls you make before you hit Generate:
- The voice: deep and measured for a documentary, warm for a story, clear for a lesson.
- The text: short sentences, one idea each, and any tricky numbers spelled out the way you want them said.
- The pacing: how fast the voice talks, the gap between sentences, and longer pauses where the story calls for them.
Which voice does each kind of narration need?
Read a ghost story in an ad voice and the tension's gone. Put a storybook voice on a software tutorial and it sounds like a joke. So figure out what you're narrating first:
| Type of narration | Voice | Pace | Tip |
|---|---|---|---|
| Documentary | Deep and calm | Unhurried | Silence between sections so the footage can breathe |
| Kids' story | Warm and expressive | Slow | Short sentences and a pause before the surprise |
| Audiobook | Natural, easy to hear for hours | Medium, steady | Same voice and same settings in every chapter |
| Explainer and e-learning | Clear and neutral | Medium | One step per sentence and a pause after each concept |
| Mystery or horror | Low and restrained | Slow, long silences | The pause before the twist is scarier than the twist |
| Advertising | Upbeat, smiling | Brisk | Short copy, brand name last |
Documentary narrator voice
A documentary voice backs up the picture instead of competing with it. Look for a deep or mid-range voice without much flourish and slow it down a notch. What really makes it sound like TV is the silence: a beat of one or two seconds between sections where the footage does the talking.
"At thirteen thousand feet, the air holds half the oxygen." That's how documentary copy reads: present tense, fact up front. Bury the fact at the end of a clause and the listener has tuned out by the time it arrives.
How do you narrate a bedtime story?
Warmly, and without rushing. Pick a close, friendly voice, set it a little slower than normal and widen the gap between sentences. For a bedtime story, dropping the speed does more than any sound effect.
If characters talk, dialogue mode gives each one its own voice (more on that below). Stick to generic or original characters, and to your own stories or public-domain ones.
Audiobook narrator AI
A very distinctive voice is charming for a minute and exhausting by the hour. So pick an audiobook voice by running a full page of the book through it and listening for at least five minutes, never with a single test line.
Lock in speed, sentence spacing and volume, and leave them alone between chapters. If chapter 4 runs faster than chapter 3, listeners pick up on it right away.
Explainer videos and e-learning
Here the voice works for the screen. Clear, normal speed, with a pause after each concept so people have time to look. If the course has several modules, don't switch voices: it ends up being the voice of your course.
Mystery, horror and true-crime style stories
Fear lives in silence. A Dramatic-style voice or a low, restrained one, slowed down, with a long pause right before the line that matters: "She opened the door. <#2#> No one was there." Two seconds of nothing beat any adjective.
Ads and promos
A radio or social ad lasts seconds. Energetic voice, speed a bit above normal, short gaps between sentences. Write less than you think will fit, and close with the brand and what you want the listener to do.
How to write a script for narration
Listeners can't scroll back. Every sentence has to land the first time, whether the narrator is a person or an AI:
- If you'd need a breath halfway through a sentence, split it. A period is a pause; a comma is half of one.
- One idea per sentence, with the key word at the end, where the stress naturally falls.
- Question and exclamation marks change the intonation, and an ellipsis stretches the pause. Use them on purpose.
- Ordinary amounts, dates and times read fine. Anything that can be said two ways, like "1/2", "3x" or an acronym you don't spell out, write in words exactly as you want to hear it.
- Cut long parentheticals, footnotes, links and lists of more than four items. They get lost out loud.
- Read it aloud yourself before generating. Wherever you stumble, the listener gets lost.
Custom pauses with <#1.5#>
For a silence longer than a period gives you, put a tag with the seconds between two sentences: <#1.5#> leaves a second and a half, <#3#> leaves three, up to 10 per tag. Use it to change scenes in a story, let a documentary shot breathe, or build tension.
Tags aren't read out or charged: only the characters that get spoken count.
That night, the lighthouse went dark for the first time in a hundred years. <#2#>
No one in town could explain why. <#1#>
No one, except the keeper.Setting the narrator's pace
The generator has two controls. Speed, from 0.5× to 2×, sets how fast the voice talks. Space between sentences, from Tight to Loose (up to 800 ms), sets how much silence sits between one sentence and the next. You can preview the voice with those settings before generating and save them to it for next time.
To start: documentaries and audiobooks at normal speed or just under, stories and horror slower with more space, ads a touch faster with less. Fine-tune with a real paragraph of your script.
For a human benchmark, the National Center for Voice and Speech puts average conversational American English at about 150 words per minute. Narration usually runs a little slower than a fast conversation.
Create AI narration in Voizum, step by step
Once the script is written:
- 1Open the voice library and filter by language and style. Start with Narration; for a documentary also try Informative, and for horror, Dramatic.
- 2Play the samples of three or four voices and keep the one that fits the table above. There are 300+ voices in 7 languages: English, Spanish, German, French, Portuguese, Italian and Russian.
- 3In the generator, paste the script or click Upload a document (it accepts .txt, .md, .srt, .pdf, .docx). The language is detected automatically.
- 4Drop <#1.5#> tags wherever you want a long silence.
- 5Set the speed and sentence spacing and preview the voice with them before generating. If you like them, save them to the voice.
- 6Click Generate. You see the credit cost before anything is spent.
- 7Listen to the result and download the MP3 for your video editor or podcast host.
Can I use my own voice as the AI narrator voice?
Yes, and if you run a channel, a course or a podcast, it's probably your best option. From a 15 to 60 second recording, Voizum builds a copy of your voice; from then on you write the script and your voice narrates it without recording again.
Record the sample the way you want the narration to sound. Rush through it in a casual phone-video tone and the clone will narrate the same way. Cloning costs 199 credits and unlocks with your first credit purchase.
To narrate in another language, accent-free mode copies only your voice's timbre. And if you publish on YouTube, its help page says cloning your own voice to create voiceovers or dubs doesn't require the altered or synthetic content label.
Stories with characters and dialogue mode
In dialogue mode you mark who speaks at the start of each line and everything comes out as one audio file, with up to 5 characters and 100 turns. Paste the whole script in one go:
How to format the script
Put the name in square brackets, in <>, or followed by a colon. Lines below without a name stay with that character, so a long paragraph doesn't need the tag repeated. Then assign voices: a calm one for the narrator, and character voices that are easy to tell apart.
[Narrator] The fox hadn't eaten in three days when he spotted the crow on the branch.
[Fox] What shiny feathers! I bet your voice is even lovelier.
[Crow] Do you really think so?
[Narrator] And the crow, full of pride, opened its beak.Settings that help
Keep "Balance the voices with each other" on, or it'll sound like separate recordings glued together. Expressiveness goes from Restrained to Lively: bump it up for a kids' story and leave it in the middle for a serious piece.
Audiobooks and long-form narration
A single audio holds up to 600,000 characters, about 10 hours of speech, so a short book fits in one go. Still, one file per chapter works better for an audiobook: listeners can skip around and you can change chapter 7 without touching the rest.
With Chain audios you generate up to 100 audio files at once, each with its own name and a single minimum charge for the whole batch. Like cloning, it unlocks with your first credit purchase.
Before publishing an AI-narrated audiobook, check the store's rules, because they vary a lot:
- ACX (Audible's production platform) requires audiobooks to be narrated by a human unless otherwise authorized, and prohibits unauthorized text to speech or AI narration.
- Google Play Books runs its own auto-narrated audiobook program, provided you hold the audiobook rights to the title.
- Only narrate text you wrote, have rights to, or that's in the public domain.
What it costs
Sign in with Google and you get 600 free credits every month, about 10 minutes of narration, enough to test several voices on your own script. They renew on their own, work in the generator and dialogue mode, and no card is needed.
After that, every 1,000 characters cost 60 credits, about 60 per minute of audio. No subscription, and purchased credits never expire.