Translate a video and keep the same voice
Your narrator, speaking Japanese, German or Hindi
Subtitles ask the viewer to read. A stock AI voice sounds like a different person. Seed Audio 1.5 re-voices the speech in another language while keeping the voice of the person who spoke, and the translated track comes back as long as the original, so it drops straight onto the picture.
Works from English or Chinese into 30 languages.
Hear it before you try it
A 9-minute English film, translated into 12 languages with Seed Audio 1.5 by our team. Play it and switch the language in Settings → Audio track. Listen for the same narrator in every version.
- Source
- English
- Target languages
- 12
- Requests
- 24
- Length difference
- < 0.03 s
Who this is for
YouTube channels
One video, many audio tracks. Viewers in Tokyo and Madrid hear the same host in their own language without a second upload.
Courses and explainers
A teacher's voice is part of the lesson. Translating it keeps the person, not just the words.
Product videos and ads
Launch a demo in a new market the same day, with the presenter your customers already know.
Documentaries and stories
Narrated films, like our Pompeii test, are the easiest case: one voice, clear pauses, steady pace.
How to translate a video in the same voice
This is the exact order we followed for the film above.
Check the source language
Video translation starts from English or Chinese. The model can detect which one it is, so you only choose the target language.
Cut long videos at pauses
One request takes 4 to 360 seconds of video. Our 9-minute film went in as two parts, 4:43 and 4:17, each cut where the narrator stops for breath. A cut mid-sentence is audible in the translation.
List names and rare words
Proper names and old or technical words are where translations slip. Add them to the glossary of the request so they come out the same in every language.
Run one language at a time
Each request returns one language. In our runs a part of about four and a half minutes took between 157 and 200 seconds.
Listen before you publish
Play every language once, at least the opening and the names. Fixing a word before upload is far cheaper than after viewers find it.
Put the tracks back
The translated track is the length of the original, so it lines up with the picture as it is. Publish separate videos per language, or add every track to one YouTube video.
What changes and what stays
| Part of the video | After translation |
|---|---|
| Spoken words | Translated into the target language |
| Voice character | Kept: the same speaker, not a stock voice |
| Background music | Kept |
| Sound effects | Kept |
| Length | Same as the original (in our test, under 0.03 s apart) |
| Picture and lip movement | Unchanged: this is audio translation, not lip sync |
Music and effects are kept according to ByteDance's documentation. In our own test we translated the narration on its own and mixed the music back ourselves.
Translate into any of 30 languages
From English or Chinese into:
Spanish and Portuguese each come in two variants. Pick the one your audience speaks: Mexican or Castilian Spanish, Brazilian or European Portuguese.
Limits to know
- The source must be English or Chinese. A video in Spanish cannot be translated directly.
- One request covers 4 to 360 seconds of video, up to 200 MB, in 480p, 720p or 1080p.
- Names can change form between languages unless you list them in the glossary.
- The mouths on screen are not redrawn. On close-ups of a talking face the mismatch shows; on narration it does not.
- Translate only videos and voices you have the rights to.
Questions about same-voice translation
Can AI translate a video and keep the original voice?
Yes. Seed Audio 1.5 video translation re-voices the speech in the target language while keeping the speaker's voice character. In our test, the same narrator was recognisable across all 12 languages.
Will the background music be lost?
According to ByteDance, video translation keeps the original music and sound effects and replaces only the speech. In our own test we sent the narration without music and mixed it back ourselves, which is also a safe route if your project files have the music on its own track.
Does the translated audio match the length of the video?
In our test, yes: each part came back less than 0.03 seconds shorter than the source, in every language, so no retiming was needed.
Can I translate a video longer than 6 minutes?
Yes, in parts. One request takes up to 360 seconds, so split the video at natural pauses, translate each part and join the results. Our 9-minute film went in as two parts.
Is this lip sync?
No. The voice is translated; the picture stays as it is. For narration, screen recordings and voice-over videos that is all you need.
Which languages can I translate into?
Thirty, including Spanish (Spain and Mexico), Portuguese (Brazil and Portugal), Japanese, Korean, German, French, Italian, Russian, Hindi, Arabic, Turkish and Vietnamese. The source has to be English or Chinese.
Keep going
Your voice, in the next language
Translate one short clip first and listen to the names. Then do the whole video.
Translate a video