Glossary

Who this is for: everyone.

Terms used across this wiki, in alphabetical order.

Admin (event producer)

The role that runs a captioning session: operates the capture page and editor, manages rooms and tokens, and prepares overlays and transcripts.

abra

The command-line tool from Co-op Cloud used to deploy and manage recipes. Run it on the machine that talks to your server.

Bearer token

The way a program proves it may use the speech service's API: it sends Authorization: Bearer <token>, where the token is the service access token. See Advanced API.

Capture page

The browser page where you choose an audio input, start captioning, and read the live text.

Compute type

A low-level setting that trades speed for accuracy or memory use (for example int8 or float16). Most users can leave it on the default.

Co-op Cloud

A self-hosting toolset that deploys packaged "recipes" with abra. This stack ships as the caption-ninja recipe, the recommended way to run the services.

Docker Compose

The tool that builds and runs the whole stack from the captions-deploy repository on one machine. Best for development, evaluation and testing.

Editor

A page where a human corrects the live captions before they are displayed.

GPU

A graphics card. Speeds up speech recognition; optional but recommended for live use.

Hoster (sysadmin)

The role that deploys and maintains the stack: installs the services, manages the domain and TLS, controls access, and takes backups.

Model

The trained data file the speech service uses. Bigger is more accurate but slower and hungrier for memory. Common choices: base, small, medium, large-v3.

Origin

The scheme, host and port a web page is loaded from, e.g. https://caption.example.org. The relay only accepts connections from origins you list, with no trailing slash.

Overlay

A page showing only caption text, meant to be dropped into OBS, a stream, or a projector.

Relay

A WebSocket message forwarder that moves caption text between browsers. It does not perform speech recognition.

Room

A named channel on the relay, such as source or output. Rooms are isolated from one another.

Room token

A password for one room, scoped to read or write.

Service access token (CAPTION_API_KEY)

A password protecting the speech service's API. Optional in trusted local use.

Speech service (caption-local)

The program that performs speech recognition (speech to text) and can serve the capture page.

TLS

The encryption behind https:// and wss://. Needed for any public hosting so tokens and captions are not sent in the clear.

Transcript

The saved text of what was captured, downloadable as a file.

User

The role that watches or reads captions. Users open a link in a browser and do not install or run anything.

Trusted origin / allowed origin

An origin explicitly permitted to reach a service. Configured in the speech service and relay; anything else is refused.

WebSocket

A browser connection that stays open so text can flow both ways in real time. The relay and the capture page use it.

Whisper

The open-source speech recognition technology used by the speech service, by OpenAI. "Whisper" refers to the engine, not to this project.

Write / Read token

See Room token.