Local voice typing · Windows · macOS · Linux

Speak the way you speak. It types what you meant.

Hold a button, talk, let go. Your words are transcribed on your own machine, and a small local model fixes broken grammar and misheard words before they reach the cursor, in any app.

Works offline No account ~2.5 s to paste
# release· any app with a cursor
A
Arjun10:38

Staging build is green. Shipping after lunch?

M
Mei10:41

Is the login crash fixed yet?

Message #release
a simulation, no microphone used

Wherever there’s a cursor

English fixer

Other dictation types what it heard. This types what you meant.

The speech model writes down exactly what it hears, mistakes and all. A second model on your machine reads the whole sentence and rewrites it into clear English. Flip the switch to compare.

Showing: what gets pasted
Grammar

I already tell him many time but he still never fix the bug.

I’ve already told him many times, but he still hasn’t fixed the bug.

Misheard

I want to deploy the model on the GP you so the inference more fast.

I want to deploy the model on the GPU so that inference is faster.

Fixed, not obeyed

Write a email to my boss say I am sick today.

Write an email to my boss saying I am sick today.

Trimmed

…you all please don’t late.

Please don’t be late.

Two hearings

Where you mumbled, it knows.

Each clip is transcribed twice in one GPU call, with and without your accent note. Where they disagree, the fixer sees both. Cost: 0.15 s.

A send me the in voice by Friday
B send me the invoice by Friday
Recent context

The topic settles the word.

Your last two dictations from the past three minutes ride along, so a sound-alike resolves to what you’ve been talking about.

earlier the script crashes on import
now    pie tone → Python
Guardrails

It won’t put words in your mouth.

If a rewrite looks wrong in any of these ways, it’s thrown away and your raw transcript is pasted instead. One- and two-word dictations are never rewritten.

  • changes a number
  • grows past 1.6×
  • shrinks below 0.5×
  • keeps <60% of your words
  • answers instead of fixing
  • adds paragraphs
  • translates
  • repeats the last dictation
corrections.txt

Teach it your repeat mistakes.

data bass => database
cube control => kubectl
Tray

A traffic light is the whole UI.

Ready
Working
Paused
Any shortcut

Mouse side button, or record your own.

Ctrl+Alt+J
Clipboard history

Left-click the light for your last five copies.

Ship the build after lunch.Call 5551234 before five.It costs 3200 dollars.
Models

We tested twelve. You get the two best.

Every candidate ran 28 weak-English sentences: grammar, misheard words, context, traps that must be fixed rather than answered, sentences to leave alone, and numbers. Each was scored on the text that would actually be pasted. Switch between the winners from the tray.

google/gemma-4-E2B-itbestFixed “pie tone” and “in voice” from context, every trap
28/281.44 s
Qwen/Qwen3-4B-Instruct-2507Twice the time for one more case
27/282.03 s
Qwen/Qwen3-1.7Bdefault27/28 with corrections.txt; never changed a meaning
26/281.02 s
Qwen/Qwen2.5-1.5B-Instruct“Please don’t be late” became “Please arrive early”
26/280.75 s
SmolLM2-1.7B-InstructEdits sentences that were already right
24/281.03 s
Qwen/Qwen3-0.6BLeaves most misheard words alone
21/280.63 s
openbmb/MiniCPM5-1BAnswers questions instead of fixing them
17/280.49 s

Measured 2026-11-16 on an Intel Arc 130V, bf16, greedy decoding, without corrections.txt. The prompt was tuned on Qwen, so the comparison favours it a little. The full table of 12 and the eval script are in the README.

On-device

Nothing leaves your machine. It’s still fast.

Let go → pasted
2.5 s

For a spoken sentence on a Core Ultra 5 laptop with its built-in Arc GPU. Transcription and the English fix take about a second each.

  • No upload
  • No account
  • No telemetry
  • HF_HUB_OFFLINE=1
Where the time goes
Arc GPU≈ 2.5 s
CPU only5–7 s + fixer
transcribe fix English paste CPU transcribe CPU fixer: not measured

NVIDIA, Apple and Intel Arc GPUs are picked up automatically, and it falls back to the CPU if the GPU fails.

Under the hood

From button to cursor in seven steps.

One dictation is processed at a time, so quick ones queue instead of fighting over the GPU. The log keeps what was heard next to what was pasted.

  1. Record 16 kHz

    Hold the shortcut. On Windows it’s swallowed, so the side button stops paging your browser back.

  2. Gate < 0.5 s dropped

    Silence is trimmed, quiet speech boosted, and clips with no speech (a fan, a knock) are dropped.

  3. Transcribe twice ~1 s

    Qwen3-ASR-1.7B, with and without your accent note, in one batched call.

  4. Clean up instant

    “one two three” becomes 123, “comma” becomes a comma, filler words go. Then corrections.txt.

  5. Fix the English ~1 s

    The local model gets both hearings plus your recent dictations as context.

  6. Check instant

    Guardrails compare the rewrite with the transcript. When in doubt, the transcript wins.

  7. Paste Ctrl/Cmd+V

    Into whatever window has focus. The text stays on the clipboard too.

Install

Set up in five minutes. Then it’s offline for good.

PowerShell
> windows\setup.bat
# the speech model, then the English fixer (~8 GB, once)
> .venv\Scripts\python.exe download_model.py
> .venv\Scripts\python.exe download_model.py --model Qwen/Qwen3-1.7B
> windows\run_dictate_background.vbs
Python 3.10+3.12 recommended. On Windows, tick “Add python.exe to PATH”.
About 10 GB of diskSpeech model 4.4 GB, English fixer 3.4 GB, plus the environment.
A GPU helps a lotOn Intel Arc / Core Ultra, install the XPU build of torch for 6–7× faster transcription. Check with dictate.py --doctor.
First start: ~45 sWhen the tray light turns red, hold the shortcut and talk.
Questions

The honest fine print.

Does anything get sent to the internet?

No. The models are downloaded once from Hugging Face. After that the app runs with HF_HUB_OFFLINE=1: no upload, no account, no telemetry. Audio and text stay on your machine.

Is it tested on macOS and Linux?

Not yet on real hardware. It was built and measured on Windows 11 (Core Ultra 5 228V). The macOS and Linux code paths are written and share the same core. On Linux you need an X11 session, because Wayland blocks global shortcuts and tray icons for ordinary apps.

Which languages work?

Speech recognition covers 52 languages and dialects, including strong Chinese and mixed Chinese-English. The rewrite step is for English only, so turn off Fix my English in the tray if you dictate in another language.

What if the fixer changes my meaning?

The guardrails reject most bad rewrites and paste your raw words instead. The log keeps what was heard next to what was pasted, so you can add repeat mistakes to corrections.txt, or switch the fixer off with one click.

Can I use Gemma 4 instead of Qwen?

Yes, from the tray. Gemma 4 E2B scored 28/28 but needs transformers 5, so it runs as a worker in a second environment (.venv-t5). If it fails to load, the app falls back to Qwen3-1.7B and says why in the log.

Why doesn’t it type into some windows?

Windows won’t let a normal app send keystrokes to one running as administrator. Use Restart as Administrator from the tray for those.

Stop typing. Start saying.

Free, local and private. Your voice and your words never leave your machine.