Murmur

Free, local voice dictation for Windows and Linux.

Hold a key, speak, release. Murmur transcribes your voice using faster-whisper and pastes the result into whatever app you have focused. Optionally, Claude Haiku cleans up filler words and fixes punctuation before pasting.

No subscription. No cloud. Your audio never leaves your machine - unless you choose to use a remote server you control yourself.

---

Features

• Push-to-talk - hold any configurable key to record, release to transcribe and paste
• Local transcription - runs Whisper entirely on your own GPU or CPU, no internet required
• Remote transcription - offload transcription to another machine over your local network, a VPN, or a reverse proxy
• AI cleanup - Claude Haiku removes filler words, fixes punctuation and capitalization, and preserves your language (Dutch, English, or mixed)
• Style profiles - choose Formal, Informal, Technical, or write your own style instruction
• User profile - tell Haiku who you are so it can apply context to every transcription
• Word corrections - force correct spelling for names, terms or brand names Whisper gets wrong; applied even without AI cleanup
• Saved servers - store multiple remote server configurations and switch between them instantly
• Audio device selection - choose any input device from settings
• Activity log - compact or debug view of every transcription session; full history saved to disk
• Whisper Server manager - install, start, stop and configure a local Whisper server directly from the tray icon (Windows); status and connection info on Linux
• System tray - runs silently in the background, waveform icon changes colour for idle / recording / processing
• Single installer - one .exe sets up a Python virtual environment and all dependencies automatically (Windows); setup_linux.sh for Linux

---

Requirements

Windows

Dependency	Version	Download
Python	3.10 or newer	python.org/downloads

⚠ During Python installation, tick "Add Python to PATH".

Linux (Ubuntu / Debian / Zorin OS)

Run setup_linux.sh — it handles everything automatically. See Linux setup below.

For GPU transcription (recommended)

Dependency	Version	Download
NVIDIA GPU Driver	Latest	nvidia.com/drivers
NVIDIA CUDA Toolkit	12.x	developer.nvidia.com/cuda-toolkit-archive

Without CUDA, Murmur falls back to CPU transcription automatically. The installer detects this and sets a lighter model (medium, int8) as the default.

For AI cleanup (optional)

Dependency	Notes	Link
Anthropic API key	Free tier available	console.anthropic.com

---

Installation

Windows

1. Download the latest installer from Releases
2. Run Murmur-Setup-vX.X.exe
3. The installer will:
   • Create a Python virtual environment
   • Install all Python dependencies
   • Create a Start Menu shortcut (and optionally a desktop shortcut)
4. On first launch, Murmur downloads the Whisper speech model (~300 MB for medium, ~1.5 GB for large-v3). A loading screen will appear - just wait until it disappears.

Linux setup

git clone https://github.com/R3PC0N/murmur.git
cd murmur
bash setup_linux.sh

The script installs all system packages, creates a virtual environment, installs Python dependencies, enables the AppIndicator tray extension, and adds your user to the input group for global hotkeys.

⚠ After setup, log out and back in once so the hotkey takes effect (input group).

Start Murmur:

./murmur.sh

---

First-run setup

1. AI cleanup (optional but recommended)

Windows: open %LOCALAPPDATA%\Murmur\.env in a text editor and add your key:

ANTHROPIC_API_KEY=sk-ant-your-key-here

Linux: open .env in the Murmur folder and add the same line.

Or enter it directly in Settings → AI Cleanup.

2. Start dictating

Murmur starts in the system tray. Hold F9 anywhere, speak, release. The transcribed text is pasted into your active window.

The hotkey can be changed in Settings → General.

---

Settings overview

Open Settings from the tray icon (right-click → Settings).

Section	What you can configure
General	Push-to-talk key, start with system
Audio	Input device
Transcription	Local or remote mode, Whisper model and device, saved remote servers
AI Cleanup	Enable/disable Claude Haiku, Anthropic API key
Display	Recording overlay, sound feedback
Profile	Transcription style, user context, word corrections

Word corrections

In Profile → Word corrections, add one correction per line:

murmur=Murmur
cuda=CUDA

Corrections are applied after transcription using whole-word matching. They work even when AI cleanup is disabled.

Style profiles

Choose a style in Profile → Transcription style:

Style	Effect
none	No style instruction - only filler removal and punctuation fixes
formal	Complete sentences, professional tone
informal	Casual tone, contractions allowed
technical	Technical terms and acronyms preserved exactly
custom	Write your own instruction

---

Remote transcription

Murmur can send audio to a Whisper server running on another machine - useful if your laptop is slow but you have a powerful desktop or home server.

How it works

1. The server runs a FastAPI service that accepts audio and returns transcribed text
2. Communication is secured with an API key
3. The client sends a small WAV file over HTTP and receives the transcribed text back

All you need is for the client to be able to reach the server's URL. How you arrange that is up to you.

Connecting client to server

There are several ways to make the server reachable from another device:

Local network — if both devices are on the same Wi-Fi or LAN, use the server's local IP directly:

http://192.168.1.x:8765

Tailscale — a free zero-config VPN. Install it on both devices, sign in with the same account, and use the server's 100.x.x.x Tailscale IP:

http://100.x.x.x:8765

If you use a VPN (e.g. Mullvad), add Tailscale to its split-tunnel exclusions so both can run simultaneously.

Reverse proxy — if you run Caddy, Nginx, or a similar proxy, add a virtual host that forwards to port 8765. This lets you use a domain name with HTTPS and works from any network without installing extra software:

https://whisper.yourdomain.com

Direct port forwarding — open port 8765 (or 443 via a reverse proxy) on your router and point it at the server machine. Combine with a dynamic DNS service if your home IP changes.

Windows server (via Murmur UI)

If your server is a Windows PC with Murmur installed:

1. On the server PC: right-click the Murmur tray icon → Whisper Server...
2. Click Install Server (one-time setup, downloads ~500 MB)
3. Click Generate to create an API key
4. Note the Remote URL and API key shown in the window
5. On the client: open Settings → Transcription, switch to Remote, paste the URL and key
6. Click Save current as... to save this server for quick access later

The server can be started and stopped from the tray icon at any time. Enable "Start server when Murmur launches" to have it start automatically.

Linux server (Ubuntu / Debian)

Requirements: Python 3.10+, CUDA 12.x (for GPU), or CPU-only.

cd murmur/server
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
nano .env  # set MURMUR_API_KEY and optionally WHISPER_MODEL, WHISPER_DEVICE

.env example:

MURMUR_API_KEY=your-strong-random-key
WHISPER_MODEL=medium
WHISPER_DEVICE=cuda
WHISPER_COMPUTE_TYPE=float16

Run manually:

python faster_whisper_server.py

Run as a systemd service:

sudo cp murmur-whisper.service /etc/systemd/system/
# Edit the service file to match your paths and username
sudo systemctl enable --now murmur-whisper

If your CUDA libraries are in a non-standard location (e.g. installed via Ollama), add the path to LD_LIBRARY_PATH in the service file:

Environment="LD_LIBRARY_PATH=/usr/local/lib/ollama/cuda_v12"

Verify the server is running:

curl http://localhost:8765/health
# {"status":"ok","model":"medium","device":"cuda"}

---

Activity log

Right-click the tray icon → Activity log to see recent transcriptions.

• Compact - shows transcription results and errors only
• Debug - shows every step including raw Whisper output before cleanup

Full history is saved to disk. Open it via tray → Open history.

---

Building from source

Windows:

git clone https://github.com/R3PC0N/murmur.git
cd murmur
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
python main.py

Linux:

git clone https://github.com/R3PC0N/murmur.git
cd murmur
bash setup_linux.sh
./murmur.sh

Building the Windows installer (requires Inno Setup 6):

Open murmur.iss in Inno Setup → press F9
Output: dist\Murmur-Setup-vX.X.exe

---

Privacy

• Audio is processed locally by default and never sent anywhere
• When using remote mode, audio is sent over HTTP to a server you control — secure it with HTTPS (via a reverse proxy) or a VPN if used over the internet
• AI cleanup sends transcribed text (not audio) to the Anthropic API if enabled

---

License

MIT - see LICENSE