mirror of
https://github.com/prdlk/noctalia-ostt.git
synced 2026-08-02 09:31:38 +00:00
90 lines
3.7 KiB
Markdown
90 lines
3.7 KiB
Markdown
# OSTT
|
|
|
|
Voice-to-text for the Noctalia bar, powered by
|
|
[OSTT](https://github.com/kristoferlund/ostt).
|
|
|
|
Click the microphone (or bind a hotkey) to start recording — the capsule turns
|
|
into a red record indicator with an elapsed timer while the microphone is
|
|
live. Click again to stop: OSTT transcribes the audio and, by default,
|
|
**copies the transcript to your clipboard** (`ostt -c`) — ready to paste.
|
|
No popup terminal is spawned; the recorder runs headlessly and the bar capsule
|
|
is the only UI.
|
|
|
|
The indicator also lights up for recordings started outside Noctalia (e.g. an
|
|
`ostt launch` hotkey), since the plugin watches OSTT's own runtime state.
|
|
|
|
## Plugin
|
|
|
|
| Field | Value |
|
|
| --- | --- |
|
|
| ID | `prdlk/ostt` |
|
|
| Entries | Bar widget: `recorder`; service: `monitor` |
|
|
| Requires | [`ostt`](https://ostt.ai/guide/installation), `script` (util-linux), `wl-clipboard` on Wayland (`xclip` on X11) |
|
|
|
|
Pick a transcription model once with `ostt model` / `ostt auth`.
|
|
|
|
### Output modes
|
|
|
|
**Clipboard (default)** — `ostt -c` copies the transcript with `wl-copy`
|
|
(Wayland) or `xclip` (X11). Paste it anywhere with your normal paste; the
|
|
transcript is also readable from scripts via `wl-paste`.
|
|
|
|
**Paste (opt-in)** — `ostt --paste` copies the transcript, then injects a
|
|
single paste shortcut (`wtype`/`ydotool`/`xdotool`) into the focused app. Per
|
|
the [OSTT docs](https://ostt.ai/guide/configuration#paste-output) it sends
|
|
exactly one configurable key — default `ctrl+v` — and cannot detect success,
|
|
so it silently does nothing in apps with a different paste binding (terminals
|
|
usually want `ctrl+shift+v`). Tune `[output.paste] paste_key` in
|
|
`~/.config/ostt/ostt.toml` before enabling it.
|
|
|
|
## Usage
|
|
|
|
Add the **OSTT** widget from the Add-widget picker.
|
|
|
|
| Interaction | Action |
|
|
| --- | --- |
|
|
| Left click | Start recording / stop & transcribe |
|
|
| Right click | Cancel the in-flight session (no transcription) |
|
|
|
|
Widget states: mic glyph (idle) → red filled mic + `m:ss` timer (recording) →
|
|
spinner (transcribing).
|
|
|
|
## Settings
|
|
|
|
| Setting | Type | Default | Description |
|
|
| --- | --- | --- | --- |
|
|
| `output_mode` | `select` | `clipboard` | Where the transcript goes: copy to the clipboard, or paste into the focused app. |
|
|
| `model` | `string` | *(empty)* | Optional `provider/model` override (e.g. `deepgram/nova-3`, `whisper/turbo`). Empty uses the model selected with `ostt model`. |
|
|
| `process_action` | `string` | *(empty)* | Optional post-processing action id from `ostt.toml` (`ostt -p <action>`). |
|
|
| `show_elapsed` | `bool` | `true` | Show the elapsed timer while recording. |
|
|
| `notify_on_done` | `bool` | `true` | Notify when the transcript has been delivered. |
|
|
| `extra_args` | `string` | *(empty)* | Advanced: appended verbatim to `ostt record` (e.g. `--param language=sv`). |
|
|
|
|
## Hotkeys (IPC)
|
|
|
|
The widget exposes IPC events for compositor keybinds:
|
|
|
|
```sh
|
|
noctalia msg plugin prdlk/ostt:recorder focused toggle # start / stop & transcribe
|
|
noctalia msg plugin prdlk/ostt:recorder focused start
|
|
noctalia msg plugin prdlk/ostt:recorder focused stop
|
|
noctalia msg plugin prdlk/ostt:recorder focused cancel
|
|
```
|
|
|
|
Example (niri):
|
|
|
|
```kdl
|
|
Mod+Space { spawn "noctalia" "msg" "plugin" "prdlk/ostt:recorder" "focused" "toggle"; }
|
|
```
|
|
|
|
## How it works
|
|
|
|
- OSTT's recorder is a TUI, so the widget runs it inside a throwaway pty via
|
|
`script(1)` — headless, no window.
|
|
- Stopping uses OSTT's external-trigger contract: `SIGUSR1` finishes the
|
|
recording and runs transcription (the same mechanism `ostt launch` uses).
|
|
- The `monitor` service polls `$XDG_RUNTIME_DIR/ostt/recording.pid`
|
|
(liveness-checked) and publishes `idle / starting / recording / transcribing`
|
|
to the widget, plus an exit-code file written by the pty wrapper so failed
|
|
transcriptions surface as error notifications.
|