How to help

Draft: awaiting approval

This text is a draft. It is not yet approved. The open questions below are for the operator to answer before it is published.

Download: the link, the checksum and the notarisation note, once the app is released (D-7).
What the app does: the defaults below (idle start, battery, download caps) are the volunteer plan's; confirm them against the released app.
The volunteer agreement (D-12): step 3 now links its draft; its open questions are listed on that page.
Who may volunteer: invite-only, people known to the operator (D-3); confirm the wording.
Windows: keep the line saying it comes later?

Lend your computer's idle time to dub talks and lectures from any language into many others. LinguaForge sends your computer one short piece of a dub at a time; it records the voice takes while you are away, and the project checks the takes before they are used.

In short

Ask for an invite with the invite request form. Volunteers are invited personally, because some jobs carry a short clip of a speaker's voice.
Install the app. [download link, checksum and notarisation note]
Enrol with your invite code. You accept the volunteer agreement. One invite is for one computer.
Choose a public name, or stay anonymous. A name is lowercase letters, digits and hyphens, 3 to 24 characters, and is shown only after it is approved. Anonymous hours count in the totals only.
Set the hours when your computer may work, for example nights only, and tick the languages it accepts.

What you need

A Mac with Apple silicon (M1 or later), macOS 14 Sonoma or later. Windows: later.
16 GB of memory for OmniVoice; 24 GB for VoxCPM2; 32 GB for the largest voice model (Qwen3-TTS).
About 10 GB of free disk space.
Mains power: by default the app works only when the computer is plugged in.
A first download of 4 to 9 GB for each voice model, plus about 3 GB for the speech checker (Whisper). "All languages" is an opt-in, larger download: it adds the check models some languages need (Persian about 3 GB).

What the app does

It starts only when your computer is idle: by default after 10 minutes without input, or 2 minutes after the screen locks. A video, a call or a presentation counts as in use.
It works on mains power; on battery only if you allow it and the charge is above 80 %, and never in Low Power Mode.
It pauses when the computer gets hot or short of memory, and when you come back it stops after the take it is on. If you keep working for two minutes, it hands the piece back and frees the memory.
It runs at low priority, keeps the computer awake only while it works (you can turn that off) and downloads models only on an unmetered network, by default at most 20 MB/s and 50 GB a month.

What the app never does

It never sends your computer's name, your user name or a list of your files. It sends only the chip, memory, operating system and app version, your public name if you chose one, and the state of its work.
It never writes a speaker's voice clip to disk unencrypted, and deletes it when the job ends.
It never runs code from a job: jobs are data only, signed by the project, and checked before use.
It never downloads a model except at the versions on the app's signed list, each file checked.

Questions

Will it slow my computer down?: It works only while you are away, and when you come back it stops after the take it is on, usually within a few seconds.
How are hours counted?: Only once a dub is published: for each piece your computer made that was accepted and used in the published track, each piece once. Hours are shown to a tenth of an hour.
Which languages?: Any source language and any target language the voice models support. You choose which ones your computer accepts.
Whose voice is used?: Some dubs copy the speaker's own voice, always with the speaker's written consent; the others use a different voice. See Privacy.
Can I stop?: Yes, at any time: pause the app or remove it. To have your data deleted, see Privacy.