Meet our new collection,
handmade in Belgium,
delivered to your door in 48 hours.
What is an AI voiceover?
An AI voiceover (text-to-speech, or synthetic voice) reads your text in a human-sounding voice produced by an AI model. Modern voices breathe, vary their intonation and pause naturally, so on simple sentences they are hard to tell apart from a recording.
There are two main types:
- Stock voices: ready-made voices you pick by language, gender and tone (calm, upbeat, warm).
- Cloned voices: a model reproduces a real person's voice from a recorded sample. Useful when a trainer or founder needs to "speak" in videos they have no time to record, provided they agree to it.
Step by step
- Write the script
Use sentences of 8 to 15 words, about 70 to 80 words per 30 seconds, with the hook in the first line.
- Adapt it for speech
Write numbers, acronyms and brand names the way they should be pronounced.
- Pick and test a voice
Listen to two or three voices reading your own script and keep the one that fits your audience.
- Generate and cut picture
Generate the narration, then place your shots and B-roll over the sentences.
- Mix and caption
Duck the music under the voice, add synced captions and check the overall level.
- Disclose when needed
If the voice imitates a real person, get their consent and label the AI use when you publish.
| Video length | Word count | Typical use |
|---|---|---|
| 15 seconds | 35 to 40 words | Ad, teaser |
| 30 seconds | 70 to 80 words | Product video, quick tip |
| 60 seconds | 140 to 160 words | Short tutorial, explainer Reel |
| 90 seconds | 210 to 240 words | Service overview |
| 3 minutes | 420 to 480 words | Long Short, detailed tutorial |
When an AI voice works, and when to skip it
A synthetic voice is a good fit for:
- tutorials and screen recordings, where the picture matters more than the presenter;
- product videos built from B-roll, with nobody on camera;
- other-language versions of an existing video;
- pickups: replacing one mispronounced line without reshooting;
- high-volume series such as weekly tips or industry news.
It is a weaker choice for testimonials, videos where your personality is the point, and emotional content. On social media, a real voice with small imperfections often earns more trust than polished narration. A sensible split: your own voice for talking-head videos, an AI voice for explainers.
How to write a script for text-to-speech
The AI voice reads exactly what you write. Text written to be read on screen sounds stiff when spoken. A few rules:
- Short sentences: 8 to 15 words, one idea each.
- Pace: narration usually runs at 130 to 160 words per minute. Plan about 70 to 80 words for 30 seconds and 140 to 160 words for a minute.
- Punctuation: a comma gives a short pause, a period a longer one. Ellipses and exclamation marks change the intonation, so test them.
- Numbers and acronyms: write "twenty-four dollars" rather than "$24", and "S-E-O" if the voice reads "SEO" as one word.
- Brand names: if a name comes out wrong, spell it phonetically in the voice script. Your captions can keep the correct spelling.
- Hook: the first 2 to 3 seconds, roughly 6 to 10 words, should tell viewers what they will get.
Example for a 20-second product video (about 50 words): "Lose your keys every morning? This key finder rings the moment you call it. One battery lasts a year. It fits in your pocket, it is water resistant, and setup takes two minutes. Link in the description."
Choosing the right voice
- Language and accent: American, British, Australian or Indian English each feel different to an audience. Match the accent to your viewers.
- Tone: calm for a tutorial, energetic for an ad, warm for a local business. Listen to the voice on your own script, not only the sample.
- Consistency: keep the same voice across videos. It becomes part of your brand, like a jingle.
- Speed: most tools let you speed the voice up. Beyond 10 to 15% faster than normal, it starts to sound unnatural.
Editing and mixing the voiceover
- Cut picture to voice, not the other way around: generate the narration first, then lay shots over the sentences.
- Duck the music: keep it well below the voice, roughly 15 to 20 dB lower while someone speaks, and let it come back up in the gaps.
- Add captions: many viewers watch with the sound off. Captions synced to the narration remain essential.
- Leave gaps: 0.3 to 0.5 seconds of silence between ideas makes narration sound more human.
- Output level: aim for an even level with no peaks; around -14 LUFS is a common target for online video.
Voice cloning, consent and disclosure
Never clone someone's voice without their written consent, even a colleague's or a celebrity's "as a joke". A voice is personal data and part of a person's identity.
In the European Union, Article 50 of the EU AI Act has applied since 2 August 2026: AI-generated or manipulated audio or video that realistically resembles an existing person (a deepfake) must be disclosed as such. Platforms add their own rules. YouTube asks creators to disclose realistic altered or synthetic content, audio included, at upload (official YouTube help), and TikTok asks creators to label realistic AI-generated content.
In practice, a stock voice reading your tutorial script is not an issue. A cloned voice that makes a real person say realistic things needs that person's permission and a clear label. When in doubt, add "AI-generated voice" to the description.
Making an AI voiceover with Orbicut
In Orbicut, AI voiceover uses credits: 10 credits per 30 seconds of narration, from the 30 credits on the free plan to the 300 a month on Starter. You write or paste the script, choose a voice, and the narration lands on the timeline with its animated captions. Music ducks automatically under the voice. For multi-language versions, pair it with project translation (25 credits, 5 languages). Plan details are on the pricing page.
Frequently asked questions
Do AI voiceovers sound natural?
On short, well-punctuated sentences, modern synthetic voices sound very natural. Problems mostly show up on brand names, acronyms and long sentences, which you fix in the script.
How many words is a 30-second voiceover?
About 70 to 80 words, based on a typical narration pace of 130 to 160 words per minute.
Is AI voice cloning legal?
Cloning your own voice, or the voice of someone who has consented, is allowed. In the EU, the AI Act has required since 2 August 2026 that content realistically imitating an existing person be disclosed.
Do I have to disclose an AI voiceover on YouTube?
YouTube requires disclosure of realistic altered or synthetic content, including audio, that viewers could mistake for real, such as a cloned voice putting words in a real person's mouth. Generic narration over a tutorial usually does not fall into that category.
Does Orbicut offer voice cloning?
No. Orbicut offers ready-made AI voices, billed at 10 credits per 30 seconds on every plan. Cloning your own voice is not an announced feature.