Free & open source · Local · On-device

Murmur

Speak. Murmur writes it back clean.

Tap a key, talk, and your words land in any app — fillers dropped, your names spelled right, sentences capitalized, spelling fixed. Then Murmur tells you the one or two words to double-check, so you fix those instead of re-reading everything. Every other tool dumps the raw transcript. Murmur corrects it, on your machine. No cloud, no account — free and open source. Windows.

listening
You said
↓ rules · your dictionary · spell-check
Murmur wrote
// a live demo of the on-device cleanup — no text leaves this page
01 / Capture

Tap once. Talk. Tap again.

Hands-free by default — no holding a key down. Add push-to-talk too if you like; every trigger is live at once.
Right Ctrl

Tap to start / stop

The default. Tap once and talk as long as you want, tap again to insert. A key that does nothing else on Windows.

Hands-free
F9 (optional)

Push-to-talk

Prefer walkie-talkie? Set a hold key: hold it, talk, release to insert. Runs alongside the tap toggle.

Hold
double / triple-tap

Your call

Or toggle with a double- or triple-tap, or a combo like Ctrl+Shift. Pick what your hands like in ⚙ Settings.

Configurable
02 / Cleanup

It doesn't just transcribe. It cleans.

Layers run in order, each free and on-device. The cloud tools polish on their servers — Murmur does it on yours, then stops.
01

Rules

Capitalizes sentences, drops um / uh / you know, turns "new line" into a real break, tidies spacing.

● local · instant
02

Your dictionary

Fixes your names & jargon — get max → Getmax, shri ram → Sriram. Learns new ones from your edits.

● local · instant
03

Spell-check

Offline spelling correction that protects your proper nouns — it never "fixes" your names into something else.

● local · offline
04

Smart AI

Optional grammar & homophone pass (their/there) via a local Ollama model. Lights up when you pull one — still 100% offline.

○ optional · local LLM
A cloud option (Claude Haiku) exists, off by default — the only setting that ever sends text off your machine.
03 / Verify

It tells you which words to double-check.

The slow part of dictation isn't the talking — it's hunting for the one word it got wrong. Murmur already knows: it scores every word and surfaces only the suspect ones.
verify · press F8

Send the deck to Priyanka before their standup — I think she'll recieve it by noon.

"recieve" — click to fix
↳ Replace → received Edit this word… It's correct — don't flag again

Solid underline — the recognizer was unsure, or it's a word Murmur has never seen.

Dashed underline — a homophone (their / there): heard fine, just worth a one-second glance.

Click it, pick the fix, and Murmur learns it — next time you say that word, it's already right.

Nobody else does this. Cloud tools hand you a transcript and make you proofread the whole thing. Murmur does the proofreading triage for you — on your machine.
04 / It learns you

A dictionary that remembers your words.

The longer you use it, the less it gets wrong — without ever phoning home.

Fix it once

Correct a note in Browse and Murmur diffs your edit, remembering wrong → right for next time.

teaches itself
🎯

Heard right up front

Your words are fed to the recognizer too, so the names and jargon you use every day come out right at the source — not just patched after.

primes the model
📒

Yours to edit

A plain custom_dictionary.json you own — add names, jargon, acronyms, protected words.

open + local
🔒

Never leaves

Your dictionary and learned edits live on your disk. No account, no sync, no server sees a word.

private by design
05 / Browse

Every note, on a real calendar.

Not a flat list. A painted Month grid and a 7-column Week — click a day to dive into what you said.
Month · click a day
Week · 7 columns
Each day shows how many notes you dictated; notes auto-file into projects, and every clip is searchable with the app you said it in.
06 / On your machine

The whole thing runs on the laptop in front of you.

No servers. Nothing synced. The most private dictation can be — because nothing leaves your machine.
0
bytes leave your device by default. Speech is transcribed and cleaned in memory, on your CPU.
$0
cloud cost. Runs entirely on your machine — faster-whisper, pyspellchecker, Ollama. No servers, no per-seat bill.
100%
local. Works on a plane. Your voice, your notes, your vocabulary — yours alone.
07 / The stack

Built on open source — and what plugs in next.

Honestly tagged: what's live now, and what wakes up when you want more. Nothing faked.
🎙

Hearing

Live: faster-whisper (local Whisper, CPU) + high-pass / spectral noise reduction before transcribing.

on-device STT
🧹

Cleaning

Live: rule engine + learning dictionary + pyspellchecker. Next: deeper grammar models.

the difference
🧠

Smart pass

Auto-detected: Ollama local LLM for homophones / grammar. Haiku (cloud) opt-in only.

local-first AI
⌨️

Plumbing

Live: pynput triggers, paste-anywhere, Obsidian daily notes, tray + autostart, .wav + manifest.

zero install
FAQ

Straight answers before you ask.

The questions people actually have about a "local" dictation tool. No marketing fog.

Does my voice or text ever leave my computer?

No. By default Murmur transcribes and cleans your speech entirely on your own machine — no cloud, no account, no server sees a word. The only exception is an optional Claude Haiku polish tier that's off by default; everything else, including the optional local AI, runs fully offline.

How is it different from Wispr Flow?

Wispr Flow sends your audio to the cloud. Murmur runs 100% locally on Windows, and instead of only transcribing it cleans the text and tells you which words to double-check — so your voice never leaves your machine.

How is it different from Dragon?

Dragon is heavy, enterprise and $500+ a year. Murmur is free, installs in minutes, runs locally, and learns your names and jargon as you correct it.

Does it work offline?

Yes — on a plane, with no internet. Speech recognition (faster-whisper), the cleanup engine and spell-check all run on-device.

Which languages does it support?

English runs on a fast English-only model. Tamil, Hindi, Telugu, Malayalam, Kannada, Spanish and roughly 99 other languages are supported through Whisper's multilingual model.

Is it really free? What's the catch?

The local app is free. A future Pro tier (cloud sync, team dictionaries, profession vocabulary packs) will be paid — but the private, on-device core stays free.

Get it now

Murmur is free & open source. Download & run it today.

Windows · 100% local. No account, no cloud, nothing uploaded.

The easy way — one click, no setup.

  1. Download the Windows build and unzip it anywhere.
  2. Double-click Murmur.exe. (First launch downloads the local speech model once, ~150 MB.)
  3. Press Right Ctrl, talk, press again — your cleaned text pastes in.

Not code-signed yet, so Windows may say "Windows protected your PC" — click More info → Run anyway. Murmur runs only on your machine and uploads nothing.


Prefer to run from source? (developers)

git clone the repo → double-click install.batrun.bat. Needs Python 3.10+.

08 / Behind Murmur

Built by GetMax — an AI-native company that ships software which does the work, not just describes it.

Murmur is one tool. The same team builds production AI agents — revenue-cycle, eligibility, operations — running every day.
GETMAX

We don't just talk about automation — we run it. Agents that read, decide and act inside real healthcare operations, alongside public builds like Murmur and Sentinel. Same hands, same standard.

Visit GetMax