AI dubbing can now turn a spoken video into another language quickly, and some tools also adjust the speaker’s mouth so the new audio looks natural. It works best on clean, front-facing speech with a modest quality bar. It struggles with fast dialogue, emotional performance, side angles and languages with very different rhythm, and it raises consent and disclosure questions that subtitles avoid entirely.
What the technology actually does
Three separate things are often bundled together under “AI dubbing”:
- Translation. The script is translated, ideally by a person who understands the market, not only by a machine.
- Synthetic voice. A voice reads the translation, sometimes modelled on the original speaker’s tone.
- Lip-sync. The speaker’s mouth movements are altered to match the new audio.
The first two are mature enough for many business uses. The third is newer. YouTube, for example, describes its auto dubbing as available to everyone in 27 languages, while its lip-sync feature, which adjusts the speaker’s lip movements to match the translated audio, was described as a pilot at the time of writing (November 2026). Its separate “Expressive Speech” option, which aims to preserve the creator’s original emotion and energy, covers a smaller set of languages (YouTube Official Blog). Treat any vendor claim of “invisible” dubbing as a claim to test on your own footage.
Where quality tends to break down
These are practical limits worth checking in any test clip, rather than guarantees:
- Timing differences between languages. The same sentence can take noticeably longer or shorter in another language. The translation may need rewriting to fit, which a literal translation will not do.
- Performance. Humour, sarcasm, grief and urgency live in delivery. A synthetic voice can flatten them even when the words are correct.
- Difficult shots. Profile views, hands covering the mouth, beards, heavy shadow and fast cuts give lip-sync models less to work with.
- Names, products and jargon. Brand names and technical terms are often mispronounced unless you supply pronunciation guidance.
- Multiple speakers and overlapping speech. Separating voices and keeping each one consistent adds steps.
- Cultural fit. A correct translation can still carry the wrong register or references. This is the same gap explained in Localisation vs Translation.
Consent and disclosure
A dub puts words in a real person’s mouth, in a voice and with lip movements they did not produce. Before you do it:
- Get written agreement from every person on screen that their voice and face may be altered, and say which languages and channels it covers.
- Check contracts with presenters, actors and voice talent for rules on synthetic reuse.
- Keep the original and the altered versions so you can show what changed.
Platform rules also matter. YouTube requires creators to disclose realistic AI-generated or altered content, including content that shows a real person saying or doing something they did not. Its help page lists some edits as not needing disclosure, such as captions, upscaling and cloning your own voice for voiceovers (YouTube Help). Whether a lip-synced dub of a real person falls under that rule is a judgement to make against the current page, not an assumption. The broader picture is covered in Disclosing AI in Advertising. This is general information, not legal advice.
When subtitles are the better choice
Subtitles translate the spoken audio into text while leaving the original performance untouched. They are the safer option when:
| Situation | Better fit |
|---|---|
| A founder, expert or customer speaks on camera and authenticity matters | Subtitles |
| The speaker has not agreed to a synthetic voice | Subtitles |
| The footage has fast dialogue, profile angles or overlapping speakers | Subtitles |
| The audience is used to watching with sound off | Subtitles |
| The video is narration over visuals with no on-screen speaker | Dubbing, with no lip-sync needed |
| Animated or stylised characters | Dubbing, where lip-sync is less scrutinised |
The W3C makes a useful distinction: captions are in the same language as the audio and serve accessibility, while subtitles translate the audio into another language and are usually not an accessibility accommodation on their own (W3C WAI). If you dub, you still need same-language captions for deaf and hard of hearing viewers.
A sensible testing process
- Pick one representative 20 to 30 second clip, including one difficult shot.
- Have a fluent speaker of the target language review the translation before generating anything.
- Compare three versions: subtitles, voice dub without lip-sync, and voice dub with lip-sync.
- Ask native speakers, not the internal team, which one feels trustworthy and why.
- Decide per video type, not once for the whole library.
For campaign-wide thinking beyond language, see Adapting Video Campaigns for Different Markets and Cultures.
Where Radiance Impact fits
Radiance Impact’s Studio works on voice and sound, post-production and short-form video, and its Global team supports localisation and adaptation for English and Chinese audiences. If you are weighing dubbing against subtitles for a specific video, get in touch or see the localisation page.
Sources
- YouTube auto dubbing: expressive speech and lip sync pilot, YouTube Official Blog
- Disclosing use of altered or synthetic content, YouTube Help
- Captions/Subtitles, W3C Web Accessibility Initiative
