murmur

Record system audio from meetings (Zoom, Google Meet, Teams, etc.) on Linux using PipeWire + FFmpeg.

Why

No-fuss local meeting recording — hit a keyboard shortcut when a call starts, hit it again when it ends. Audio is saved locally, no cloud services or meeting bots needed.

Goals

• Simple toggle workflow — one shortcut key to start/stop recording, with desktop notifications
• Universal — captures any audio playing through your speakers/headphones, works with any meeting app
• Local-first — recordings stay on disk at ~/Recordings/meetings/, no uploads
• Low overhead — just FFmpeg recording a PipeWire monitor source
• Future: transcription — pipe recordings through Whisper or a transcription API for searchable meeting notes

How it works

1. Discovers PipeWire audio sinks via wpctl
2. Attaches to the monitor source of your default output device (captures everything you hear)
3. Records via FFmpeg's PulseAudio input (pipewire-pulse compatibility layer)
4. Saves audio file + JSON metadata (timestamps, source, duration)

Requirements

• Linux with PipeWire + pipewire-pulse
• FFmpeg
• wpctl (WirePlumber)
• notify-send (for desktop notifications from toggle)
• Python 3.14+ / uv

Install

# Core only (recording, toggle, watch, devices)
uv pip install murmur

# With TUI dashboard
uv pip install murmur[tui]

# With AI summarization (DSPy + LiteLLM)
uv pip install murmur[ai]

# With transcription (faster-whisper)
uv pip install murmur[transcribe]

# Everything
uv pip install murmur[all]

# Or run from the project directory
uv run murmur --help

Usage

# List audio output devices
murmur devices

# Start recording (interactive, Ctrl+C to stop)
murmur start
murmur start --tag standup --format mp3

# Toggle recording on/off (for keyboard shortcuts)
murmur toggle

# Check recording status
murmur status

# List saved recordings
murmur list

Meeting detection

Murmur can watch for meeting apps (Zoom, Google Meet, Teams, etc.) using your microphone and notify you — or automatically start recording.

# Get notified when a meeting app grabs your mic
murmur watch

# Auto-record when a meeting is detected
murmur watch --auto-record

# Auto-record with mic input (dual-channel: system left, mic right)
murmur watch --auto-record --mic

# Faster polling (default: 5s)
murmur watch --interval 3

Works by polling PipeWire for Stream/Input/Audio nodes — detects Chrome, Firefox, Zoom, Teams, Slack, Discord, WebEx, Skype, and more. When the app releases the mic, you get a second notification and auto-recording stops.

Keyboard shortcut

Super+Shift+R toggles recording on/off with a desktop notification.

Set up via GNOME custom shortcuts — runs murmur toggle in the background.

Recordings

Saved to ~/Recordings/meetings/ with naming convention:

meeting_standup_2026-03-18_14-30-00.flac
meeting_standup_2026-03-18_14-30-00.json   # metadata

Configuration

Optional TOML config at ~/.config/murmur/config.toml:

[recording]
output_dir = "~/Recordings/meetings"
format = "flac"

[watch]
interval = 3
auto_record = true
apps = ["chrome", "firefox", "zoom", "teams", "slack", "discord"]

[transcribe]
auto = true          # auto-transcribe after recording
model = "base"       # whisper model size
language = "en"

[summarize]
auto = true          # auto-summarize after transcription
model = "llama3"     # ollama model
ollama_url = "http://localhost:11434"

[diarize]
hf_token = "hf_..."  # hugging face token for pyannote

Plugins

Plugin	Command	What it does	Install
watch	murmur watch	Detect meeting apps using the mic, notify + auto-record	built-in
memory	murmur memory	Personal context for LLM summaries	built-in
tui	murmur tui	Live dashboard with artifact viewer + generation	murmur[tui]
summarize	murmur summarize <file>	DSPy structured summarization → .summary.md	murmur[ai]
transcribe	murmur transcribe <file>	Whisper transcription → .txt + .srt	murmur[transcribe]
diarize	murmur diarize <file>	Speaker diarization → .rttm + .diarized.txt	murmur[diarize]

Development

# Set up dev environment
make install

# Run all checks (lint + format + test)
make check

# Individual targets
make lint          # ruff check
make format        # ruff format (auto-fix)
make format-check  # ruff format --check (CI mode)
make test          # pytest
make fix           # ruff check --fix

Roadmap

• Automatic transcription (Whisper local / Deepgram API)
• Speaker diarization
• Auto-detect meeting apps and start recording
• Web UI for browsing/searching recordings