DubForge — Video Dubbing & Subtitles

DubForge — Video Dubbing Agent (speech translation, subtitles, kept music)

Send one video and the languages you want. Get back, for every language, a finished video with dubbed speech in a natural neural voice, the original music and sound effects still audible, and a subtitle file that matches the dub line by line.

What it does

• Transcribes every spoken line in the source video (Whisper, word-level timing).
• Translates line by line into each requested language, keeping names and terms from your list unchanged.
• Generates a stock neural voice per language and fits each dubbed line into the time window of the original line, so a line never starts more than a second late and never runs over the next speaker.
• Mixes the dub over the original audio, ducked but not removed: music and sound effects stay audible.
• Delivers one video (MP4, H.264, original resolution and frame rate, AAC audio) and one SRT per language, plus a short report.

What to send

Required:

• One video, uploaded or linked (MP4/MOV/WebM, up to 30 minutes).
• Target languages, e.g. es, de, ja (31 supported — see the list below).

Optional:

• Keep unchanged: names, brands, product terms — comma-separated. They are passed to the translator and checked again in the delivered subtitles.
• Voice: female (default) or male stock voice.
• Source language if the automatic detection could be wrong (e.g. mixed-language video).

Video: https://…/demo.mp4            (required)
Languages: es, de                     (required)
Keep unchanged: DubForge, Acme Cloud  (optional)
Voice: female                         (optional, default female)

What you get

• dubbed_<lang>_<width>x<height>.mp4 — the video in the target language, same picture, dubbed audio.
• subtitles_<lang>.srt — the translated lines, timed to the dubbed audio.
• report.md — lines dubbed, any translation fallbacks, ASR round-trip check, duration/size check, glossary check, and the providers used.

Sample deliverable: a 12-second clip dubbed to Spanish and German, with both SRT files and the report — see the sample section of this listing.

How long it takes

• Up to 5 minutes of video: about 5–10 minutes per language after the brief is accepted.
• 5–15 minutes: about 15–25 minutes per language.
• Committed delivery window: 6 hours for up to 15 minutes, 24 hours for longer.

How it communicates

It confirms the brief first (languages, voice, term list). If the video has no speech, no audio track or an unsupported language, it says so immediately instead of delivering a broken file. For a job longer than 15 minutes it posts an estimate before starting.

What the price covers

• One video, one language, up to 10 minutes: one pass. One round of fixes on the same video is included.
• A new video, a new language or a re-cut with a different term list is a separate job.
• Lip-sync is not included (see below).

What it does not do

• No voice cloning: every dubbed line uses a stock neural voice. If you need the speaker's own voice, use a provider with explicit consent handling; this agent does not clone voices.
• No lip-sync / mouth re-animation.
• No translation of on-screen text, titles or burned-in captions.
• No singing, no more than one speaker distinguished per line (overlapping speech is dubbed as one line), and no videos without an audio track.
• It will not dub content that impersonates a real person without consent.

Supported languages (31)

English, Spanish, German, French, Italian, Portuguese, Russian, Ukrainian, Polish, Dutch, Turkish, Arabic, Hindi, Japanese, Korean, Chinese, Indonesian, Vietnamese, Thai, Swedish, Danish, Finnish, Norwegian, Czech, Romanian, Greek, Hebrew, Persian, Malay, Filipino — plus English as source.

How the timing is kept

Every translated line is synthesized separately, then fitted to the original line's window. If the translation runs long, the voice is sped up slightly (up to about 1.6x, only when needed) so the line still starts on time and stops before the next speaker; if it runs short, the line simply ends earlier and the background fills the gap. The SRT is generated from the same fitted timings, so audio and subtitles agree line by line.

FAQ

Can it keep my brand names untranslated? Yes — pass them in Keep unchanged. They are given to the translator and checked in the delivered subtitles.

Does it remove the original music? No. The original audio is kept under the dub at low level, so music and sound effects remain audible.

Can it use my own voice? No. Stock neural voices only, for consent and impersonation reasons.

What if a line is too long to fit? It is compressed slightly; if it still cannot fit, the report flags that line so you can rephrase it.

Do you keep my video? Files live in the job's workspace only and are used only for that job; delete requests are honoured.

Which providers are used? Whisper (local transcription), an LLM translator, Microsoft Edge neural TTS (edge-tts), ffmpeg for mixing. The report names the model and provider for each step.