VEED Dubbing: translate a video and keep the speaker's voice

You made a good video in one language and you would like the people who speak another one to watch it too. Subtitles work, but they ask the viewer to read instead of watch. VEED Dubbing takes the other route: it listens to the speech in your video, translates it, speaks it in the new language with a voice cloned from the original speaker, and moves the lips to match. This post is the practical version: what to prepare, what to set, and what to check before you publish.

What goes in

• The video (required), up to ten minutes long, with the speech in its own audio track.
• Target language (required): a long list, from major languages to small regional ones. This model has no regional dialects; if you pass a regional tag such as French from France, it dubs in the base language.
• Source language (advanced): set to auto by default, which detects what is spoken. Set it by hand when the video has an accent, a mix of languages or a lot of background noise.
• Key terms (advanced): a short list of names and terms to leave as they are, up to ten, each short. Put the brand, the product and the speaker's name here.

Before you run it

1. 01Clean the audio. The model works from what it can hear. A voice over loud music gets mistranslated first; if you have a clean voice track, use it, or clean the sound of the video before dubbing.
2. 02Cut the video into one talk per file. Ten minutes is the ceiling, and a short file is also faster to check. Long talks are easier to review in chapters.
3. 03Write the key terms. Anything you would be annoyed to hear translated, a product name or a slogan, goes in the key terms field.
4. 04Know the language you are targeting. If you cannot read the result, find someone who can for the first minute. A dub that sounds smooth is not always correct.

A cheap trial run

What to look for in the result

• Timing. Some languages need more words than others to say the same thing. If a sentence feels rushed, the original was probably dense; trim the source speech and try again.
• Names and numbers. Check each one against your key terms and against the script.
• The voice. It is cloned from the original, so a deep voice stays deep. It will still sound a little more neutral than a person talking naturally in the new language.
• The mouth. Lip movement follows the new words. It is best on a face that stays in view; fast cuts and side profiles are where it slips.

The honest limitation

This is the standard version, and it is the one to start with, but it is not a replacement for a human translator on anything where wording matters: a legal explanation, a medical instruction, a joke that depends on a pun. If the details are critical, have a native speaker read the dub. If you need to control how closely the new voice follows the old one, or to pick a regional dialect, look at VEED Dubbing Pro, which has both. And if your video has a lot of speakers talking over each other, split the file by scene, because overlapping voices are the hardest case for any dubbing model.

Other video models, and the ones that make speaking characters from a single picture, are on the models page. To start, upload a short clip in the studio.