You can set up offline voice-to-text on your Mac in a few minutes. Download Chirp, grant two permissions, and press the Fn key to start dictating. Every word is transcribed on your Mac using whisper.cpp, so you do not need an internet connection or an account to dictate.
This post is the quick Mac setup. For the bigger picture (what on-device dictation means, what Chirp sends over the network, and how it compares with Apple Dictation, MacWhisper, Superwhisper and others), read our complete guide to private, offline dictation on Mac and Windows.
What You Need
- A Mac running macOS 13 (Ventura) or later, Apple Silicon or Intel
- About 53 MB for the download (the speech model is included)
- A working microphone (built-in or external)
- An internet connection only to download the app
Step-by-Step Setup
Step 1: Download Chirp
Go to chirp.la/mac and download the installer. There is one universal build that runs natively on both Apple Silicon and Intel Macs, so you do not need to pick a version.
Step 2: Install the .dmg
Open the downloaded .dmg and drag Chirp into your Applications folder. The app is signed and notarized by Apple, so it opens like any other app from an identified developer.
Step 3: Grant Microphone and Accessibility Permissions
Chirp needs microphone access so it can hear you, and Accessibility access so it can detect the Fn key and paste text into the app you are using. macOS may also ask whether Chirp can control System Events, which is how it sends the paste keystroke. Chirp's Permissions screen shows the status of each and opens the right page in System Settings.
Step 4: Press Fn to Start Recording
Click into any text field, then press the Fn key once. You will hear a short sound and the menu bar icon changes to show Chirp is listening. Option+M does the same thing if you prefer a key combination.
Step 5: Speak Naturally
Talk at your normal pace, in English, in complete sentences. A single recording can run for up to 5 minutes; Chirp counts down the last 20 seconds and then stops on its own.
Step 6: Press Fn Again — the Text Is Pasted
Press Fn again to stop. Chirp transcribes the audio on your Mac, deletes the recording, and pastes the text where your cursor is. The text is also left on your clipboard and saved in Chirp's local history. On Apple Silicon a short dictation usually takes well under a second.
How It Works Under the Hood
Chirp uses whisper.cpp, an open-source C/C++ implementation of OpenAI's Whisper speech model. The bundled model is ggml-tiny.en-q5_0, a quantised version of the English-only Whisper tiny model, about 30 MB. On Apple Silicon it runs on the GPU through Metal; Intel Macs use the CPU and are a little slower. The small model keeps Chirp fast and light, but it makes more mistakes than larger models with strong accents, jargon and background noise.
Tips for Better Accuracy
- Keep the microphone close. A headset or a mic near your mouth makes a bigger difference than anything else.
- Speak in complete sentences rather than isolated words. The model uses context to choose words and punctuation.
- Minimize background noise. Music, fans and other voices all cost accuracy.
- Dictate in chunks. A paragraph or two at a time is easy to review and stays well inside the 5-minute limit.
- Use your history. Every transcription is saved locally, so you can search, copy or export anything you said earlier.
Frequently Asked Questions
Does Chirp work without any internet connection?
Yes, for dictation. Transcription runs entirely on your Mac, so you can turn Wi-Fi off and keep dictating. Chirp checks for updates when it launches, and once you buy a license it checks the license about once a day; a licensed copy keeps working for up to 7 days offline.
Is there a limit per transcription?
Each recording can be up to 5 minutes long. There is no word limit within that.
What about languages other than English?
Chirp is English only. The bundled model is English-only and there is no option to switch languages or models.
How much does it cost?
Chirp includes 28 free transcriptions with no account and no time limit. After that, a $9.99/year license unlocks unlimited transcriptions on up to 5 devices.