Translated speech has to preserve a message and fit the available time. Descript’s published account of its dubbing system describes that as an engineering problem. The workflow below follows its current documentation; it is a documented guide, without a hands-on product test.[8]
Use a computer with the desktop app or a supported desktop browser. Descript does not officially support editing on phones or tablets. For the full workflow below, including correcting translated text, you need a Business or Enterprise drive. Basic dubbing is available on lower current paid plans; translation proofread is the relevant restriction here.[1][5]
1. Finish the original and protect names
Correct the source transcript, finish the visual edit and style any captions before translating. Translation creates a separate composition. Later changes to the original can require another translation and more AI credits. Start with one short, completed video so the review remains manageable.[1]
On Business and Enterprise drives, open Brand Studio from the Drive view, then Glossary and the Do Not Translate list. Add product names, acronyms or other terms that should remain as written, one per line, pressing Shift and Enter to save each entry. The list matches complete words or phrases without regard to case. It preserves terms; it is not a substitute for deciding how an ordinary word should be translated.[3]
For example, a product demonstration may need a model name left untouched while the surrounding sentence changes. Separately note the numbers and conditions that must survive translation: the price, the duration, the unit of measurement and any exception to the claim. Those become specific things for your reviewer to check.
2. Choose what the translation should optimise
Open AI Tools, select Translate under Repurpose, and choose the target language. The translation method matters before you generate the voiceover.[2]
The two translation choices
| Method | What Descript does | Use it when |
|---|---|---|
| Match timing | Adjusts phrasing to fit the original speaking time. | Natural pacing inside an existing video matters most.Source 1 |
| Direct translation | Stays closer to the wording; dubbed audio may be sped up or slowed down. | You want closer phrasing and can review the resulting pace.Source 1 |
The timing problem is real. In OpenAI’s published Descript case study, the company describes dividing speech into chunks and using syllable counts, estimated speaking rates and neighbouring context to guide translation. The system is trying to fit meaningful speech into a time slot. That explains why the most literal wording may produce the least comfortable delivery.[8]
3. Review the words, then create the voice
Descript recommends that Business and Enterprise users translate captions first, review and correct the text, then apply dubbing and optional lip sync. This sequence avoids spending credits on audio and mouth movements before the wording is settled.[4]
Have someone who understands the target language compare the translation with the source. Ask them to check the marked names, quantities and qualifications, then read it aloud at the intended pace. A sentence can sound fluent while changing a promise from possible to guaranteed. A review request that names the claims is more useful than asking whether it sounds good.
To generate the voiceover, select Dub speech, assign the speaker label to a stock voice and apply the change. Descript creates a translated composition. If you subsequently correct its text, use Regenerate to update the audio; editing the words alone does not complete that step.[2]
If you need a different voice, use the translation dropdown, open the translation’s ellipsis menu and choose Change Speakers. Re-translating an existing language can overwrite its manual edits, so distinguish changing the speaker from starting the translation again.[2]
4. Add mouth movement only if the footage suits it
Lip sync is optional and is available on current Creator plans and above. Descript recommends a single, well-lit speaker facing the camera with an unobstructed mouth. It does not support multiple speakers, multitrack Sequences or avatars. For suitable footage, the Lip-sync video option is in the Translate panel.[4]
Watch the rendered version rather than judging only from the transcript. Listen across edits, where an otherwise acceptable sentence may finish awkwardly against the next shot. If the wording must change, settle it before generating another visual version.
5. Export and check the actual file
Select the translated composition. Open Export, choose Video and Local export, then export the current composition as an MP4 using the settings you need. Check that you are exporting the translated version rather than the source composition or a temporary selection.[6]
If the destination needs a separate subtitle track, Descript can export SRT or VTT files. These differ from captions permanently displayed in the picture. Choose the format your destination accepts and check the subtitle text against the final voiceover.[7]
Play the exported file from beginning to end. Once approved, give it a filename that identifies the language and keep it beside its matching subtitle track. That makes the version you reviewed easier to identify when it is time to publish.
Sources & further reading
- Translate and dub speech overview Descript Help Center · Accessed 19 Oct 2026
- Dub speech to add translated voiceover Descript Help Center · Accessed 19 Oct 2026
- Manage your “do not translate” list Descript Help Center · Accessed 19 Oct 2026
- Make dubbed speech look authentic with lip sync Descript Help Center · Accessed 19 Oct 2026
- Descript system requirements Descript Help Center · Accessed 19 Oct 2026
- Export a video or GIF locally Descript Help Center · Accessed 19 Oct 2026
- Export subtitles Descript Help Center · Accessed 19 Oct 2026
- How Descript engineers multilingual video dubbing at scale OpenAI · 6 Mar 2026 · Accessed 19 Oct 2026