Omega-13

Captures the last 13 seconds of audio on demand, transcribes locally or via cloud, and routes the result wherever it's needed.

---

Features

• Retroactive ring buffer — continuously holds 13 seconds of JACK/PipeWire audio in memory; saving reconstructs a linear WAV from circular segments with zero re-recording
• Local or cloud transcription — routes audio to a self-hosted whisper-server (HTTP) or the Groq Whisper API; provider and credentials switch at runtime via the Settings screen
• Auto-record mode — RMS-based voice-activity detection arms recording when signal crosses a configurable dB threshold and stops automatically after a configurable silence window
• Session management — recordings accumulate in a timestamped temp session under /tmp/omega13/; the session saves to permanent storage on demand, with incremental sync for recordings added after the initial save
• Multi-destination output — transcription results can be written to a Markdown file, copied to the clipboard, typed into the active Wayland window via pynput, or appended to an Obsidian daily note
• Wayland-native IPC — the app writes a PID file and handles SIGUSR1 so external tools (compositor hotkey daemons, D-Bus clients) can trigger recording without requiring keyboard focus
• Overlap deduplication — when auto-record produces consecutive clips with shared audio, the session de-duplicates transcriptions by matching word-level suffix–prefix overlaps before appending

---

Requirements

• Linux with JACK or PipeWire-JACK bridge running
• Python 3.12+
• uv for dependency management
• For local transcription: a running whisper-server instance (default: http://localhost:8080)
• For cloud transcription: a Groq API key (set via GROQ_API_KEY environment variable)
• For Obsidian daily note integration: obsidian-cli installed and configured

---

Installation

From source (recommended)

git clone https://github.com/b08x/omega-13.git
cd omega-13
uv sync

Docker (whisper-server)

A compose.yml is included for running the whisper-server dependency:

docker compose up -d

This brings up a GPU-accelerated whisper-server on port 8080. Adjust the model in compose.yml as needed.

---

Usage

uv run omega13

On first launch, the Input Selection screen prompts for JACK port selection. The choice persists to ~/.config/omega13/config.json.

Keyboard Shortcuts

Key	Action
Space / global hotkey	Capture last 13 seconds
a	Toggle auto-record mode
t	Manually trigger transcription
c	Toggle clipboard copy
j	Toggle text injection to active window
d	Toggle Obsidian daily note output
i	Open input port selector
n	Start a new session
s	Save current session to permanent storage
p	Open transcription settings
q	Quit

The global hotkey defaults to Ctrl+Alt+Space and can be changed in ~/.config/omega13/config.json.

Examples

# Launch with default settings
uv run omega13

# Override Groq API key at runtime
GROQ_API_KEY=gsk_... uv run omega13

# Trigger a capture from an external script (e.g., a Hyprland keybind)
kill -USR1 $(cat /tmp/omega13.pid)

---

Configuration

Config lives at ~/.config/omega13/config.json and is written on first run with defaults. The Settings screen (p) handles most options at runtime; manual edits require a restart.

{
  "input_ports": ["system:capture_1", "system:capture_2"],
  "save_path": "/home/user/Recordings",
  "global_hotkey": "<ctrl>+<alt>+space",
  "transcription": {
    "provider": "local",
    "server_url": "http://localhost:8080",
    "inference_path": "/inference",
    "groq_model": "whisper-large-v3-turbo",
    "auto_transcribe": true,
    "copy_to_clipboard": false,
    "inject_to_active_window": false,
    "write_to_daily_note": false
  },
  "auto_record": {
    "enabled": false,
    "begin_threshold_db": -35.0,
    "end_threshold_db": -35.0,
    "silence_duration_seconds": 10.0
  },
  "sessions": {
    "temp_root": "/tmp/omega13",
    "default_save_location": "/home/user/Recordings",
    "auto_cleanup_days": 7
  }
}

Configuration Options

Transcription

• provider: Transcription backend — "local" (whisper-server) or "groq"
• server_url: Base URL for the local whisper-server (default: "http://localhost:8080")
• inference_path: Endpoint path on the local server (default: "/inference")
• groq_model: Groq model identifier (default: "whisper-large-v3-turbo")
• auto_transcribe: Automatically transcribe after each capture (default: true)
• copy_to_clipboard: Copy result text to clipboard after transcription (default: false)
• inject_to_active_window: Type result into the focused Wayland window (default: false)
• write_to_daily_note: Append result to the current Obsidian daily note (default: false)

Auto-Record

• begin_threshold_db: RMS level above which recording starts (default: -35.0)
• end_threshold_db: RMS level below which the silence timer starts (default: -35.0)
• silence_duration_seconds: Seconds of silence before auto-stop (default: 10.0)

Sessions

• temp_root: Working directory for in-progress sessions (default: "/tmp/omega13")
• default_save_location: Destination for saved sessions (default: ~/Recordings)
• auto_cleanup_days: Age in days before unsaved temp sessions are pruned (default: 7)

---

Session Structure

Each saved session is a directory under the configured save location:

omega13_session_2024-01-15_10-30-00/
├── session.json          # metadata: timestamps, recording list, transcriptions
├── recordings/
│   ├── 001.mp4
│   └── 002.mp4
└── transcriptions/
    ├── 001.md
    └── 002.md

Sessions accumulate incrementally — recordings added after an initial save are synced to the permanent location automatically.

---

Auto-Record Mode

Pressing a arms the auto-record state machine. From that point:

1. Signal above begin_threshold_db triggers a capture automatically
2. When the signal drops below end_threshold_db, a silence countdown begins
3. After silence_duration_seconds of continuous silence, the recording stops and (if auto_transcribe is on) transcription starts
4. The system returns to armed state, ready for the next event

Recordings with average RMS below -50 dB (near-silence false triggers) are discarded silently before transcription begins.

---

Development

uv sync
uv run pytest                          # full test suite
uv run pytest tests/test_tui_bindings.py -v   # TUI binding tests only

Tests that instantiate the full app mock omega13.app.obsidian_cli and the JACK-dependent AudioEngine to avoid hardware requirements. The tests/ directory also contains demo and integration scripts (demo_*.py, example_*.py) that require real JACK and whisper-server connections.

---

Contributing

Issues and pull requests are welcome. For significant changes, opening an issue first to discuss the approach tends to save iteration time.

---

License

MIT