Vocnu

Open source engine

The same pipeline the hosted service runs, under the AGPL-3.0: speech recognition, translation, narration timing and mixing. Run it on your own computer with your own OpenAI and ElevenLabs keys.

What it does

• Recognizes speech with faster-whisper on your GPU or CPU, helped by the subtitle tracks the file already has.
• Translates with OpenAI: one analysis of the whole transcript, translation in chunks with context, an independent review and a character budget for every line, so each one fits the time it has.
• Narrates with ElevenLabs voices. Lines that run long are recorded again, rephrased shorter and only then sped up, at most 1.08 times.
• Mixes the voice-over: the original sound dips under each line, and videos keep the original audio as a second track.
• Writes MP4, MKV, MOV, WebM or AVI with switchable subtitles, M4A, MP3, WAV, FLAC or OGG audio, and SRT, VTT and ASS subtitles.
• Spells the names of known films and series the way the public names library does.

What you need

• uv and ffmpeg. Tesseract is optional, for image subtitles.
• An NVIDIA GPU makes speech recognition many times faster; the CPU works too, slowly.
• Your own OpenAI and ElevenLabs keys. You pay the providers directly: in our runs about USD 0.05 to 0.10 per minute of speech, and a receipt shows the exact amounts.

uvx --from git+https://github.com/vocnu/engine \
  vocnu-engine translate film.mkv --to es

AGPL-3.0: you may use, change and share the engine, and if you offer it as a network service, you share your changes under the same license. Notes on a language from native speakers are among the most valuable contributions.

Command line

The npm package vocnu, for Node.js 20 or newer. It uploads a file of up to 10 GB, shows the price in credits and asks before it starts, follows the job and saves the results next to the file.

npx vocnu login
npx vocnu translate film.mkv --to es

Options

Option	What it does
--output video|audio|subtitles	The video with the voice-over (default for video files), the voice-over audio or subtitles only
--from <lang>	The spoken language, detected by default
--voice female|male	A female or a male voice
--format mp4|mkv|mov|webm|avi	Video file format; by default the upload's own, else MP4
--audio-format m4a|mp3|wav|flac|ogg	Voice-over audio format
--burn-subtitles	Subtitles burned into the picture
--out <dir>	Where the results go
--yes	Start without confirming the price
--no-wait	Start the job and return its ID

Other commands

• vocnu credits
• vocnu languages
• vocnu jobs
• vocnu status <job>
• vocnu download <job>
• vocnu cancel <job>
• vocnu logout

MCP server

Add Vocnu to Claude or any other assistant that supports MCP, then ask it to translate a video. It uploads the file, tells you the price, starts only when you agree, follows the job and saves the files.

Claude Code

claude mcp add vocnu -e VOCNU_TOKEN=vocnu_... -- npx -y vocnu mcp

Claude Desktop and other MCP clients

{
  "mcpServers": {
    "vocnu": {
      "command": "npx",
      "args": ["-y", "vocnu", "mcp"],
      "env": { "VOCNU_TOKEN": "vocnu_..." }
    }
  }
}

Then ask, for example: “Translate ~/Videos/film.mkv into Lithuanian with Vocnu.”

Tools

vocnu_upload_media

Uploads a file from your computer

vocnu_start_job

Starts a job once you agree to its price

vocnu_job_status

Shows how far a job has got

vocnu_download_results

Saves the finished files

vocnu_languages

Lists the target languages

vocnu_credits

Shows your credit balance

vocnu_list_jobs

Lists your jobs

vocnu_cancel_job

Cancels a job

API tokens

Create a token in the app under Settings, API & MCP. It is shown once: the command line saves it when you run vocnu login, or you set it as VOCNU_TOKEN.

A token can upload files, show prices, start, follow and download jobs and read your credits. It cannot pay, change your profile or manage tokens. Revoke it any time; the app shows when each token was last used.

Create a token

Open data

• The voice catalog: every voice with its language, gender and a sample, as JSON.
• The names library: how the characters, places and terms of films and series are written in each language, as JSON under CC BY-SA 4.0.

Questions and answers

Is the engine free?

Yes. It is open source under the AGPL-3.0, and you pay OpenAI and ElevenLabs only for what your own keys use. The hosted service needs no keys or setup; it runs on credits.

Do the command line and the MCP server need an account?

Yes. They use the hosted service with a personal API token: the same credits, prices and refunds as the app, and you see the price before anything is charged.

Can I use the engine in a closed product?

The AGPL-3.0 asks you to share your changes when you offer the engine as a network service. For other terms, write to support@vocnu.com.