A cooking instructor records a short English video explaining how long to roast vegetables. A Spanish version is useful, but a single mistranslated temperature could make the tutorial misleading even if every lip movement looks plausible. ElevenLabs Dubbing v2 supports editing generated dubs in its UI, including changing translated text and regenerating a selected segment. This guide organizes a review around one recipe instruction, a speaker card, a timed audio segment and a human bilingual approval. The clip cards and waveforms are explanatory artwork, not screenshots or claims that a real recording was dubbed.
Lock the recipe script and original spoken values
Save the source video and approved transcript, including the instruction “Roast at 180 degrees Celsius for 20 minutes.” Note that a translator may choose a local unit convention, but any conversion or altered time needs explicit human approval. Do not let a fluent-sounding machine sentence change a safety-critical number. Mark the one speaker as Chef, and check whether any off-camera helper also speaks in the segment. Retain the original audio and room ambience for provenance. Get the person’s permission for synthetic translation and voice likeness before uploading their image or recorded voice to a third-party system.
Distinguish current Dubbing v2 from legacy Studio
ElevenLabs’ Dubbing overview explains that Dubbing v2 is the current automatic model and supports editing transcripts and translations in the web UI, then regenerating an individual segment. Its Dubbing Studio page states the older v1 Studio is in maintenance mode with critical fixes only. A new guide should not instruct readers to create a legacy Studio project as the default. For API users, the dubbing capability notes distinguish website editing from enterprise-only editing and regeneration endpoints. Confirm plan access and any per-minute cost before uploading large projects; no universal free quota is promised here.
Review speaker detection and language assignment
After processing an authorized source, inspect the detected speaker tracks and the selected output language. Confirm that the chef’s instruction stays attributed to the chef, especially if a helper answers briefly in the background. The system may preserve voice characteristics, but speaker identity is still an editorial check. If one segment merges two voices, do not publish it merely because translation text reads well. Listen to the source at the relevant timestamp and mark segments for separate attention. Keep a table of source start, source end, speaker, original text and approved target text to avoid losing context while making focused fixes.
Correct the number in just the affected translated line
Open the translated segment that contains temperature and cooking time. Compare the target-language text to the agreed recipe, and ask a fluent reviewer to verify “180 grados Celsius durante 20 minutos” in context. Do not assume changing punctuation alone will be reflected in audio until the segment is regenerated and listened to. If a term such as roasting versus baking has a specific culinary meaning for the intended region, use one approved glossary across the video. Make one targeted edit rather than re-dubbing every scene, so the reviewer can isolate what changed and identify any new issue in pronunciation.
Regenerate the segment and check timing at its boundaries
ElevenLabs describes selecting a segment for regeneration after modifying text in Dubbing v2 through the website. Listen around the start and end of the new utterance: a phrase that is longer in Spanish may spill over a cut or run into the next speaker. Look for clipped syllables, mismatched breaths and unexpected changes in perceived voice or intensity. Do not treat the new dubbed video preview as a new high-resolution master. Its documentation recommends downloading audio for production and replacing the original video’s soundtrack to preserve the source image quality. Preserve the previous segment as a review reference where the workflow permits.
Check translation meaning separately from perceived lip sync
A convincing synthetic voice can still say the wrong thing. Have a bilingual reviewer read the source and target transcripts without video, then listen while viewing the relevant picture. Verify ingredient names, allergy statements, quantities, times, temperatures and any advice that could affect safety. Do not claim that the Dubbing model itself performs perfect visual lip synchronization: ElevenLabs says its dubbing feature does not provide lip sync as part of the dubbing workflow. If a facial mismatch is distracting, adjust editorial timing or choose a suitable cutaway rather than implying the sound was captured live in the target language.
Protect original ambience and rights in the final mix
Dubbing may retain background audio, but the rebuilt voice should be monitored in context. Listen for suddenly missing kitchen room tone, unnatural ducking under a pan clang or a dubbed word competing with the sound of a tray entering an oven. Keep the licensed music, original background track and the new dub logically separated where possible. The chef’s informed permission to represent their voice in another language matters even if the interface can replicate it technically. Add translated captions from the approved text and label the translated audio appropriately, especially when a viewer might assume the person spoke Spanish during recording.
Deliver a versioned, human-approved local-language master
Export an audio track from the approved dubbing workflow and assemble it with the original high-quality picture master in an editor. Keep the exact source transcript, approved translation, regenerated-segment list, reviewer and final date. Rewatch the entire recipe, not just the corrected temperature line, checking the next instruction for pacing changes. Verify both audio and text deliver the same safety information and that any subtitles match the final dub. A successful localized deliverable is a traceable, accurate instruction in another language, not simply a synthetic voice that resembles the original speaker. The sample timelines illustrate editorial decisions rather than real model performance.
How We Research
We read ElevenLabs’ current Dubbing overview, the legacy Dubbing Studio status and its capability documentation on October 28, 2026. We did not upload a chef recording, test translation quality or clone a voice. All cooking instructions, Spanish lines, timing bars and review marks are illustrative and should be checked by an authorized bilingual editor in real work.
Frequently Asked Questions
No. ElevenLabs describes legacy Dubbing Studio as v1-only and in maintenance mode; Dubbing v2 is current.
ElevenLabs documents editing a Dubbing v2 segment in the web UI and regenerating it individually.
No. ElevenLabs says editing and regeneration through its API are Enterprise-only; check your current plan.
No. Its dubbing documentation says lip sync is not included as part of the Dubbing feature.
Use the approved dubbed audio with the original-quality source video rather than assuming a dub preview is the master.