Cirel Voice

The world's most accurate transcription.1

Free dictation for the Mac, in 60+ languages — including the Nordic ones every other tool guesses at. Words land under 200 ms behind your voice, punctuated and spelled, in whatever app you are already in.

Free forever · Notarized by Apple

Listening
Find and save your next customer.
VOICE
Speak naturally…

The problem

You speak at 250 words a minute and type at 40.

Every mail, note and reply queues up behind your two busiest fingers. The dictation you tried before made you talk like a robot, fix its commas and stay in one language — so you went back to typing.

A stream of unpunctuated speech arriving from every direction, passing through Cirel Voice and leaving as one continuous line of finished text.Moved the delivery to Thursday. Two extra pallets, same carrier. Freight to Aalborg is still open. I'll call Mikkel before five. Sending the revised offer now. Moved the delivery to Thursday. Two extra pallets, same carrier. Freight to Aalborg is still open. I'll call Mikkel before five. Sending the revised offer now. uh somoved thedeliverytil torsdagand umtwo extrasame carrieryou knowtil Aalborgstill opencall Mikkelbefore fivesorrythe freight

The benchmark

Not our benchmark. Our result.

Measured, not claimed. On the open Pipecat benchmark the big engines make two to three times our errors — and on SuperWhisper's own published benchmark, the Mac dictation apps land at five to ten percent. Lower is better. The gap is the product.

Real-time engines · open Pipecat benchmark · lower is better

Cirel Voicecirel realtime1.27%the baseline
Googlegemini transcribe live1.8× the errors2.24%
OpenAIrealtime whisper2.1× the errors2.73%
OpenAIgpt-4o transcribe2.4× the errors3.06%
ElevenLabsscribe v2 realtime2.5× the errors3.12%

Mac dictation apps · SuperWhisper's own benchmark

SuperWhispers1 voice · their best cloud model5.4× the errors6.8%
Wispr Flowno published wer · superwhisper's suite6.9× the errors8.7%
SuperWhisperwhisper large v3 · offline10.2× the errors13.0%

Time to text

<200 ms

behind your voice while you speak — and the final, corrected text is there 260 ms after your last word. Finished text, median:

Cirel Voice260 ms
Google458 ms
OpenAI740 ms

1. Among real-time engines in the open Pipecat STT benchmark, August 2026 — 1,000 real-world samples, pooled semantic WER: only the errors that change meaning, totalled across the whole set. One enterprise engine edges ahead by 0.13 points; it costs eight times as much per hour and takes four times as long to hand your words back. The app rows are plain word error rate from SuperWhisper's own published benchmark, August 2026: their strongest cloud model, their shipping offline Whisper, and Wispr Flow measured on the same suite — Wispr Flow publishes no word error rate of its own. The app rows' ×-figures compare their plain rate with our semantic one: different rulers, same direction.

Try it now

See it hear you.

This is the app's own recording panel, live in your browser. Give it your microphone and watch the grid move to your voice — nothing is recorded, and nothing leaves the page.

The magic caret

Text lands where you're looking.

No window to switch to, nothing to paste. Hold the key, talk, and the words appear at your cursor — finished, punctuated, and in the app you were already in. The gold caret is where your voice is writing.

Dictated · Mail · 0:04

Any app on your Mac

Mail, Slack, your CRM, a code editor — if it has a cursor, it takes dictation.

Punctuation as you speak

Commas, casing, paragraphs and question marks arrive with the words. Nothing to go back and fix.

Names and numbers, spelled right

Phone numbers, addresses, order IDs and Nguyễn — structured detail lands character-perfect.

What it does

Built for the way you actually talk.

Real speech is fast, multilingual and full of names the dictionary has never met. Everything below exists because that is what lands on the microphone.

Faster than your fingers

Words appear while you are still saying the sentence, and the final, corrected text is there 260 milliseconds after you stop. You never wait for the machine — it waits for you.

60+ languages, one ear

Danish, Swedish, Norwegian, English and fifty-six more, at native-speaker accuracy — accents, dialects and phone-quality audio included. No language menu; it hears which one you chose.

Switch mid-sentence

Start the sentence in Danish and finish it in English, the way you actually talk to customers abroad. One unified model follows the switch as it happens — no toggles, no pause.

Every character right

Phone numbers, addresses, order IDs, serials and names that trip up every other tool land character-perfect. Structured detail is where dictation usually dies; here it is a strength.

Knows your vocabulary

Teach it your products, your customers and your industry's terms, and recognition is guided toward the words you actually use — so jargon comes out as jargon, not as guesses.

Never stored

Audio streams encrypted, is processed in memory and is never written to disk. Processing runs on European infrastructure, in line with the rest of the Cirel platform.

How it works

Hold, talk, done.

  1. 01

    Hold the key

    One hotkey, anywhere on your Mac — mid-mail, mid-chat, mid-line of code. The gold caret marks where your words will land.

  2. 02

    Say it the way you'd say it

    Full speed, any language, several in one breath. Names, numbers and punctuation are part of the sentence, not commands you have to learn.

  3. 03

    Let go

    The finished text is already sitting at your cursor — punctuated, cased, spelled — a quarter of a second later. Read it, send it, keep moving.

Stop typing what you can say.

  1. Hold the key

    In whatever app you are already in.

  2. Talk

    Any of 60+ languages. Or two at once.

  3. Let go

    It was written before you finished.

Frequently asked questions

Everything you need to know. Can't find an answer? Talk to our team.

A dictation app for macOS. You hold a key and talk; finished, punctuated text appears at your cursor in whatever app you are in — Mail, Slack, your CRM, your editor. It runs on the same speech engine as the rest of the Cirel platform.