Private, Offline Dictation on Mac and Windows: The Complete Guide

On-Device vs Cloud Dictation

Every dictation tool does the same basic job: it records your voice and runs it through a speech recognition model that turns sound into text. The privacy question is simply where that model runs.

• Cloud dictation sends your audio over the internet to a company's servers, which transcribe it and send the text back. Cloud services can run very large models and often add AI clean-up, but your recordings leave your computer and you rely on the provider's policies about storage, review and training.
• On-device (local) dictation runs the model on your own processor. The audio never needs to leave your machine, it works without an internet connection, and there is no server-side copy to be breached, subpoenaed or reviewed. The trade-off is that the model has to fit on your computer, so it is usually smaller than what a data centre can run.

“Offline” and “private” are related but not identical. An app can transcribe locally and still send analytics, sync your transcripts to a cloud account, or send text to an AI service for formatting. Some apps offer both local and cloud models, and which one you get depends on your settings or plan. When privacy matters, look at the whole path your words take, not just the transcription step.

How to check whether a dictation tool is really on-device

1. Turn off Wi-Fi and try it. Real on-device transcription keeps working.
2. Look for a model download. Local apps ship or download a speech model (tens of megabytes to several gigabytes).
3. Read the data page, not the homepage. Look for statements such as “audio is processed on the device” or “transcription occurs in the cloud”.
4. Check which features need the cloud. AI rewriting, translation and “smart formatting” often use a server even when basic transcription is local.
5. Notice whether an account is required. Accounts often mean something is stored server-side.

How Chirp Works

Chirp is a small desktop app built with Tauri 2 and Rust. It is designed to do one thing: turn a few sentences of speech into text in whatever app you are using, without anything leaving your computer. Here is exactly what happens when you dictate:

1. You press the shortcut. Fn (or Option+M) on a Mac, Alt+M on Windows. It is a toggle: press once to start, press again to stop. A start sound plays and the tray icon changes.
2. Chirp records to a temporary file on your computer, for up to 5 minutes.
3. whisper.cpp transcribes it locally. whisper.cpp is an open-source (MIT) C/C++ implementation of OpenAI's Whisper. Chirp bundles the ggml-tiny.en-q5_0 model: the English-only “tiny” Whisper model (39 million parameters), quantised to about 30 MB. On Apple Silicon Macs it runs on the GPU via Metal, so a short dictation typically takes well under a second; Intel Macs and Windows PCs use the CPU and take a little longer.
4. The audio file is deleted as soon as transcription finishes, whether it succeeded or not.
5. The text goes where you need it. On macOS, Chirp copies it to the clipboard and pastes it at your cursor (this needs the Accessibility permission). On Windows it is copied to the clipboard and you press Ctrl+V.
6. The text is saved to a local history (an SQLite database in Chirp's app data folder), which you can search, export to CSV, or delete entry by entry.

What Chirp sends over the network

Being precise matters here, so this is the complete list:

• Transcription: nothing. It never uses the network.
• Update check: on launch, Chirp fetches a small version file from its download host and offers updates.
• License: only after the 28 free transcriptions, if you buy a license. Chirp sends the license key and a hashed device ID to activate it, then re-checks about once a day. It keeps working for up to 7 days without a successful check.
• Analytics, telemetry, crash reports: none. There is no account.

Details are in the privacy policy.

Requirements

• Mac: macOS 13 Ventura or later, Apple Silicon or Intel. One universal, signed and notarized .dmg of about 53 MB. Permissions: Microphone and Accessibility.
• Windows: Windows 10 or 11, 64-bit (x64). Signed .exe installer of about 33 MB, installs per user without admin rights.
• A microphone. A headset or a mic close to your mouth gives noticeably better results.

Limitations to Know Before You Choose

• English only. There is no language setting.
• One small model. Fast and light, but less accurate than large models on strong accents, specialist vocabulary, names and noisy rooms. There is no custom vocabulary.
• 5-minute cap per recording. Good for messages, notes and paragraphs; not a meeting recorder.
• No voice commands or AI rewriting. Chirp gives you what you said, with Whisper's punctuation.
• No automatic paste on Windows yet. Press Ctrl+V after each dictation.
• Desktop only. No Linux, iPhone or Android app.

The Alternatives, Honestly Compared

There are good tools in this space, and several will suit some people better than Chirp. Facts below come from each vendor's own pages as of September 2026; prices and features change, so check before you buy.

Apple Dictation (macOS)

Free, built in and good. Apple says Keyboard settings show whether your dictation is processed on the device; when it isn't, audio goes to Apple's servers and, according to Apple's privacy page, isn't stored unless you turn on “Improve Siri & Dictation”. It supports many languages and stops after 30 seconds of silence. If your language is processed on-device on your Mac, it is a strong free option. Chirp is for people who want local processing regardless of setup, a history of what they said, or the same tool on Windows. Chirp vs macOS Dictation.

Windows voice typing and Voice Access

Win+H voice typing is convenient, but by default it uses Microsoft's online speech recognition. Voice Access on Windows 11 (22H2 and later) works on the device and adds full voice control of the PC, which makes it excellent for accessibility. Chirp is simpler: one shortcut, local transcription on Windows 10 or 11, and a searchable history.

MacWhisper

MacWhisper is a polished Mac app for transcribing audio and video files locally, with a choice of models including larger, more accurate ones; its Pro version also offers system-wide dictation. If you transcribe recordings, interviews or podcasts, it is excellent. Chirp is lighter and only does live dictation, but also runs on Windows. Chirp vs MacWhisper.

Superwhisper

Superwhisper offers local Whisper models on its free tier and adds cloud models, custom modes and AI processing on Pro. It runs on Mac, Windows and mobile. If you want to choose bigger models or have AI reformat your text, it is a great pick. Chirp has no cloud mode at all and nothing to configure. Chirp vs Superwhisper.

FluidVoice

FluidVoice is free and open source, runs several local models (including Parakeet and Whisper), and types into the focused app. It needs macOS 15 or later on the Mac. If you like open-source tools and want model choice, try it.

Wispr Flow (cloud)

Wispr Flow is one of the most polished dictation apps, with strong AI clean-up, many languages and mobile apps. Its own data controls page states that transcription always happens in the cloud, under zero-retention agreements with its AI providers. If you are comfortable with that, it is excellent. If your requirement is that audio never leaves your computer, it is not the right tool. Chirp vs Wispr Flow.

Dragon Professional

Dragon is installed software for Windows 10 and 11 with deep voice commands and custom vocabulary, long favoured in legal and medical work. Dragon for Mac was discontinued in 2018. It costs far more than the other tools here but does far more. Chirp vs Dragon.

Otter.ai (cloud)

Otter is a cloud meeting assistant: it joins calls, identifies speakers and writes summaries. It solves a different problem from dictating into apps, and your recordings are processed and stored in its cloud. Chirp vs Otter.ai.

Which Should You Choose?

• You want dictation that never leaves your computer, on Mac and Windows, with nothing to set up: Chirp.
• You dictate in a language other than English: Apple Dictation, Superwhisper, MacWhisper or Wispr Flow.
• You need the highest accuracy on jargon or accents and have a fast Mac: an app that runs larger local models, such as MacWhisper, Superwhisper or FluidVoice.
• You want AI to rewrite and format as you speak, and cloud processing is fine: Wispr Flow or Superwhisper Pro.
• You need to control your whole PC by voice: Windows Voice Access or Dragon.
• You want meeting notes: Otter.ai or MacWhisper.

Set Up Chirp in Two Minutes

Mac

1. Download from chirp.la/mac and drag Chirp to Applications.
2. Open it and allow Microphone and Accessibility when asked (macOS may also ask to let Chirp control System Events for pasting).
3. Click into any text field, press Fn, speak, press Fn again.

Windows

1. Download from chirp.la/windows and run the installer.
2. Allow microphone access when Windows asks.
3. Press Alt+M, speak, press Alt+M again, then Ctrl+V where you want the text.

Tips: speak in full sentences, keep the microphone close, and dictate in chunks of a paragraph or two. More in our Mac setup guide and the round-up of offline speech-to-text apps.

FAQ

Last reviewed: September 25, 2026. Chirp details are checked against the current app. Details about other products come from their official pages on that date and may have changed.