AI dubbing went mainstream this year. What still needs a person?
Translation got cheap. Getting it right still takes judgment.
In February, YouTube opened automatic dubbing to all creators, with support for 27 languages. A more expressive voice mode is available in eight of them, and lip sync is being tested. Not long ago, dubbing a video into five languages was a production project. Now it is a setting.
For businesses, that changes the economics of video. A product demo, a training module or a campaign reel can reach Arabic, Hindi, Spanish and French speakers for a fraction of what it cost to hire voice talent for each language.
What the machine does well
Modern dubbing handles the mechanical work better than most people expect: transcription, translation, a voice that keeps the speaker's tone, and timing that fits the original edit. For explainers, tutorials and internal training, the result is often ready to publish after a careful listen.
What still needs a person
- Names, product terms and acronyms. Models translate things that should never be translated.
- Humour, idioms and cultural references. A literal translation of a joke is usually not a joke.
- Claims. A price, a guarantee or a regulatory statement must say exactly the same thing in every language.
- Tone for the market. Gulf Arabic, Egyptian Arabic and Modern Standard Arabic land differently with different audiences.
The practical setup is a glossary of terms that must not change, a native speaker who reviews each language before it goes out, and a clear label on dubbed content. YouTube already labels auto-dubbed videos in its player, and under the EU AI Act, AI-generated audio has to be marked as well.
Where Voxbee fits
Localization is one of Voxbee's four studios, covering dubbing, subtitles and lip sync, and it has been used on 622 videos so far. As with everything in Voxbee, a dubbed video goes out only after the customer approves it.