murmr

---

The free, local alternative to Wispr Flow - no cloud, no subscription.

murmr is a local voice dictation app for Windows. Press a key, speak, paste - no cloud, no subscription. Everything runs on your machine: transcription, AI cleanup, and optional Notion logging.

---

Demo

▶️ Watch the detailed walkthrough on YouTube

---

Why I Built This

I've been dictating more than typing - simply because it's faster. But apps like Wispr Flow send your audio to the cloud and sit behind a paywall. The built-in voice-to-text in apps like Claude and Gemini is finicky and locked to one place. I wanted something that works in every app, with nothing leaving my machine. The Notion logging came from wanting to capture thoughts as they come by just speaking, with everything automatically saved to a database I can search later.

---

Features

• Hotkey-driven - Ctrl+Win to toggle hands-free recording; hold Ctrl+Alt+Win for push-to-talk
• 100% local transcription - powered by faster-whisper; your audio never leaves your machine, zero cost per use
• Parallel transcription - silence detection splits speech into segments mid-recording; background threads transcribe each chunk as you speak, so only the final segment waits on stop
• AI cleanup - optional pass to strip fillers, fix grammar, and handle self-corrections via OpenAI or a local Ollama model
• Floating dock - draggable always-on-top pill showing live status; collapses to stay out of your way; position persists across restarts
• System tray - quiet background presence with a right-click menu for quick toggles
• Notion logging - optionally appends every transcription to a Notion page with a timestamp
• Settings GUI - full configuration window; no .env editing required

---

Architecture

Hotkey press
    └── recorder.py       starts mic stream, emits silence-split segments to queue
         └── main.py      segment worker transcribes each chunk in a background thread
              └── transcriber.py    faster-whisper inference
                   └── ai_cleaner.py    (optional) filler/grammar cleanup via OpenAI or Ollama
                        └── main.py    paste at cursor via clipboard + Ctrl+V
                             └── notion_writer.py    (optional) append to Notion page

On stop, only the final (unfinished) segment remains to be transcribed - everything before it is already done.

---

Tech Stack

Tool	Purpose
faster-whisper	Local speech-to-text inference - runs entirely on your machine
pynput	Global hotkey listener - works across all Windows apps
sounddevice	Mic capture + live RMS level for the animated waveform
pyperclip	Clipboard copy/paste
pystray	System tray icon and menu
tkinter	Floating overlay, dock widget, and settings window
Pillow	Draws the tray icon "M" lettermark in memory - no icon file needed
python-dotenv	Loads settings from .env
numpy	Raw audio buffer handling
openai	AI cleanup via OpenAI API (optional)
ollama	AI cleanup fully on-device via a local model (optional)
notion-client	Appends transcriptions to a Notion page (optional)

---

Folder Structure

murmr/
├── murmr.bat               - launcher (double-click; no terminal window)
├── README.md
└── src/
    ├── main.py             - orchestration: tray, overlay, recording flow
    ├── dock.py             - floating dock widget
    ├── hotkeys.py          - global hotkey listener (pynput)
    ├── recorder.py         - mic capture + silence-based segment splitting
    ├── transcriber.py      - faster-whisper wrapper
    ├── ai_cleaner.py       - AI cleanup (OpenAI + Ollama)
    ├── notion_writer.py    - Notion page appender
    ├── settings_window.py  - settings GUI + .env read/write
    ├── config.py           - loads .env, exports all settings
    ├── .env                - your secrets and preferences (never commit this)
    └── requirements.txt    - Python dependencies

---

Getting Started

Prerequisites

• Windows 10 or 11
• Python 3.10+
• A microphone

Clone and install

git clone https://github.com/v1jaysundaram/murmr.git
cd murmr
python -m venv .venv
.venv\Scripts\activate
pip install -r src/requirements.txt

Configure

Create src\.env and set at minimum:

WHISPER_MODEL=small   # tiny | base | small

Everything else is optional - AI cleanup, Notion logging, and overlay theme are all configurable in the Settings window.

Run

cd src
python main.py

Or just double-click murmr.bat - no terminal window.

First run: faster-whisper downloads model weights (~500 MB for small). This only happens once.

---

Usage

Action	How
Start recording	Ctrl+Win
Stop and paste	Ctrl+Win again
Push-to-talk	Hold Ctrl+Alt+Win, release to transcribe
Toggle Notion logging	Click [N] on dock, or tray → Notion logging
Toggle AI cleanup	Click [AI] on dock, or tray → AI cleanup
Open settings	Click [⚙] on dock, or tray → Settings
Collapse dock	Click [–] on dock
Quit	Click [✕] on dock, or tray - Quit

---

License

MIT © Vijay Sundaram Mohana

---

Connect

If you find murmr useful, a ⭐ on the repo goes a long way.

YouTube  ·  LinkedIn  ·  X