Decide what the audience will receive
A training video may need translated captions while retaining the original speaker. A product demonstration may need a new voice track and translated on-screen labels. A documentary interview may need careful attribution and a clearly presented translation. The same source video can therefore require different workflows depending on its audience.
Write the delivery specification first: target language and locale, subtitle format, voice requirements, video dimensions and distribution platform. Include whether the original audio must remain available. This prevents the team from buying a dubbing plan when a reviewed subtitle track would solve the actual problem.
Follow the layers, even when the pipeline is automatic
- Source understanding
- Identify speakers, terminology and the meaning of the original performance.
- Translation
- Adapt the message for the target language and context.
- Timing
- Fit speech or captions to the available time without losing essential meaning.
- Voice and sound
- Produce intelligible speech while handling music and other audio.
- Visual adaptation
- Review subtitles, on-screen text and any optional lip-sync changes.
An automated service can combine these stages, but the review questions remain distinct. A fluent voice may deliver an inaccurate translation; a correct sentence may be too long for a subtitle; a convincing mouth movement may distract from an incorrect product term.
Match automation to the type of material
ElevenLabs' current Dubbing v2 overview describes an audio-to-audio approach conditioned on the source performance, alongside automated translation and synchronisation. It also distinguishes self-service creative workflows from a production service with human translators, casting and mixing. Check the current workflow rather than assuming all dubbing operates through the same editable transcript process.
For a recurring software tutorial, build a glossary of product names, menu labels and phrases that should remain unchanged. For marketing content, review cultural references and the force of the claim. A literal translation can be grammatically acceptable while communicating the wrong promise or tone.
Keep timing from quietly changing the message
Languages take different amounts of time to express an idea. Review whether a translated line has been shortened, spoken unusually quickly or moved away from the relevant visual. For instructional material, the reference to a button should still occur when that button is visible.
Rask's official platform description includes translation controls, speaker controls, timestamps, captions and lip sync. Evaluate these as separate editing requirements. Lip sync changes visual timing; it does not establish translation accuracy, permission to use a person's likeness or the suitability of the final message.
Test a short section with a difficult term, two speakers and a pause that carries meaning. Check the complete exported segment, including background sound. A preview that sounds clear on headphones may reveal different problems when mixed with music or played on a phone.
Subtitles need their own editorial pass
HappyScribe's current platform combines transcription, subtitles and translation, with optional human services. It lists text and subtitle exports including SRT and VTT. A subtitle workflow can therefore be useful even when a new voice track is unnecessary.
Inspect line breaks, reading speed, speaker changes and whether captions obscure important content. Distinguish a translated subtitle from an accessibility caption: the latter may also need relevant sound and speaker information. Check the intended platform's requirements and whether subtitles remain selectable or are permanently burned into the video.
Voice permission and approval should travel with the asset
Confirm rights to the source recording and permission for any cloned voice or altered likeness. Do not assume that access to a recording authorises a synthetic performance. Agree on the languages, purpose and distribution, and make the use of synthetic media clear where appropriate.
Keep the approved source, translation, voice version and final export together. A later edit to a product claim may require changes in several languages. Assign a reviewer who understands the target language and subject, particularly for technical, contractual or health-related material.
Compare the complete cost of a language version
Check how source minutes, target languages, regeneration, voice features, lip sync and human review are billed. One source minute translated into several languages can create several outputs. Confirm whether application credits and API usage share an allowance and whether higher-quality exports add another cost.
A useful pilot is a finished, reviewed segment in the actual delivery format. Evaluate the time needed to correct terminology and timing, rather than choosing solely by the number of supported languages.
Localisation FAQs
Does dubbing translate text visible inside the video?
Not automatically. On-screen labels, slides and graphics may need a separate translation and editing pass.
Is lip sync necessary for every video?
No. Tutorials, narration and off-camera speakers may not need it. Choose it only when it helps the intended experience.
Can a cloned voice be used without permission?
Do not assume so. Obtain appropriate rights and consent for the intended synthetic performance and distribution.
Who should approve the final translation?
Someone competent in the target language and subject. Listen and watch the final export, not only the written translation.
Research checked 17 October 2026 against official product pages. Workflow, language and export entitlements remain product- and plan-specific.
This guide uses AI-assisted research and drafting, with source checks and editorial review.
