AI Dubbing

What happens inside an AI dubbing job

Subtitles reach people who are willing to read. Dubbing reaches everyone else — viewers on phones, children, people watching while cooking, and audiences in markets where dubbed content is the norm. Traditional dubbing means a translator, a studio and voice actors for every language, which puts it out of reach for most creators, teachers and small companies. AI dubbing compresses that pipeline into a few minutes, and this tool shows you every step so you can trust the result.

First, the speech is transcribed with a timestamp for every word, and the words are grouped into sentences. Each sentence keeps the exact moment it was spoken. Second, the sentences are translated in batches by a language model instructed to sound natural when spoken and to keep each line about as long to say as the original, because a translation that is twice as long cannot fit the same moment. Third, each translated line is voiced and placed at its original timestamp. If a line would overrun the gap before the next sentence, it is re-spoken slightly faster — never more than 1.3× — and if it still runs long, the next line is pushed back a little rather than making anyone sound rushed. Finally, the lines are mixed into one track and two subtitle files are written: one in the new language, timed to the dub, and one with the original transcript.

Keep the original voice

The most striking AI dubbing results keep the speaker's own voice. Choose 'keep the original voice' and the first two minutes of your upload are used to create a clone of the speaker, which then speaks the translation. Viewers hear the same person, now in Japanese or Spanish. Because a clone is a likeness of a real person, you must confirm that you are the speaker or have their permission; cloning adds 30 credits to the job. If the original speaker can't give consent, choose a studio voice instead — each target language has native voices, and you can preview them before starting.

Video in, dubbed video out

For video uploads, the AI dubbing tool can rebuild the video for you. After the dub is ready, 'Download dubbed video' copies your original video stream untouched into a new MP4 and replaces its audio with the dub, entirely in your browser — no re-encoding of the picture, no quality loss and nothing uploaded. If your browser can't do it, download the MP3 and SRT files and combine them in any video editor.

Where AI dubbing earns its keep

Course creators publish the same lesson in several languages without re-recording. YouTubers test a Spanish or Portuguese channel with their best videos before investing in human dubbing. Companies localize product demos and training videos. Families send a message to relatives in their language, and language learners hear the same clip in two languages side by side. For transcripts alone, use audio to text; to write and voice a new script from scratch, the AI voice generator has the same voices.

Best results from AI dubbing

• Use clips with one main speaker and little background music; music and crosstalk lower transcription and clone quality.
• Speak at a natural pace in the original. Very fast speech leaves little room for longer languages like Spanish or German.
• For 'keep the original voice', the first two minutes should contain clear speech from that speaker.
• Check names and technical terms in the translated SRT before publishing.
• Short clips first: try a one-minute section to choose the voice before dubbing a ten-minute video.
• Add the translated SRT as captions too — some viewers like both.

Current limits of AI dubbing here

One voice per dub

Each job uses one voice for all lines. For interviews with two speakers, dub each speaker's section separately or use a studio voice for both.

Up to 10 minutes

Longer videos can be split into parts and dubbed separately; the timestamps in each part start from zero.

Lip sync

Lines are timed to when each sentence starts, not to mouth movements. It works best for narration, tutorials and talking-head videos.

Refunds and recovery

If a dub fails, the credits are returned automatically. If you close the page, the job keeps running and appears when you come back.