Open source · Local-first · Higgs Audio, ElevenLabs and more
Your video, in any language, in your own voice.
Upload a video in one language and choose another. OpenDub separates the speech from the music, transcribes it, translates each sentence to fit its moment, and speaks it back in a clone of the original voice — then burns in subtitles in the new language.
- Open source. Every step is plain code you can read, run and replace.
- Local, except the voice models. Separation, speech detection, timing, mixing and rendering run on your device. Your video never leaves it — only speech clips and text go to the voice model you choose.
- Powerful. Your own cloned voice in 17 languages, each line timed to your lips and carrying your tone — surprise, enthusiasm, calm.
Other settings Cloned voice · tone matched per line
Speaking speed is not a setting: it is chosen per line so each one starts and ends with the speaker.
Replace the voice keeps the music and effects and swaps the speech. Voice-over leaves the original faintly underneath.
Dubbing
—
What it is doing
Done
—
Drag a dubbed block to move it, or its edges to change its length. Double-click to snap it back onto the original.
Lines
Change any translation, tone or timing and apply just those lines — the rest is kept.
Dub your own video
Or run OpenDub on your own machine.
The card at the top dubs videos up to five minutes right in this tab. The app on your computer has no length limit, removes the voice faster, and lets you edit, re-tone or drag any line and re-dub it in about ten seconds.
Higgs STT, translation and a cloned voice with tone tags. The best quality, and what made the dub above.
Boson AI's usage prices
Whisper, a local language model and OmniVoice. Nothing leaves your computer and nothing is charged.
Coming soon
- Install Python 3.12 and ffmpeg (Homebrew's
ffmpeghas everything needed). - Add a key for the Higgs route: put
BOSON_API_KEY=…in a.envfile. - Start it with
./run.shand openhttp://127.0.0.1:8910.
The source release is coming soon.
How it works
Four models, each doing the one thing it is best at.
- 01
Separate
Demucs splits the speech from the music and effects, so the soundtrack survives the dub and every model below hears a clean voice.
- 02
Listen
Silero VAD finds the speech. Higgs STT writes down the words in whatever language they are spoken; a local Whisper supplies only the timing, aligned word by word.
- 03
Translate to fit
The Higgs chat model translates whole sentences, each sized to how long the speaker took to say it — so the new voice starts and stops with their lips.
- 04
Speak, then check
Higgs TTS clones the voice from a clean sentence of the original and delivers each line in the tone it was said in, read from the words and from how loud and fast the speaker was. Each line is transcribed back and regenerated if it drifted, then re-said or gently stretched to match the original length.