AI dubbing — timing and culture, not technology

The duration problem

The same content takes a different amount of time in different languages, which breaks sync with picture. Three options: speed the audio (sounds wrong), extend the picture (dead air), or rewrite the script to length. Production work usually picks the third.

Lip sync

Sync features exist and still read as off on close-up speakers. For narration-led video, over-shoulder framing or screen-heavy content it does not matter. For interviews, subtitles are the safer choice — the artefacts are more distracting than reading.

Specify pronunciation

Always declare how your company, product and people's names are said. Unspecified, the model guesses from spelling, and non-English names in English audio break badly. Register the main terms in the pronunciation dictionary once.

Content that does not survive translation

• Explanations built on local regulation
• Examples that assume local business custom
• Pricing, with currency and tax treatment
• Calendar assumptions: fiscal years, seasons, holidays

Order of operations

Translate the script as text first and fix the length there; have a native reader check it if you can; then generate audio; then conform to picture; then watch it through. Starting from audio means translation errors are found after they are voiced.

FAQ