Skip to content
Whisper Local
On-device dictation and transcription for Mac

Just speak, and it becomes an AI prompt.

Press ⌥End and talk: Apple Intelligence, built into the Mac, removes fillers and false starts and types the text at the cursor of your AI chat or mail. Speech recognition and the AI both run on this Mac, so what you say is never sent anywhere.

The app is in Japanese.

The app interface is available in Japanese only.

  • Version 0.3.28
  • 5.1 MB
  • Notarized by Apple
  • macOS 14 or later
  • Apple silicon (M1 or later)
The Whisper Local result window, showing a Japanese transcript with Copy and Save as TXT buttons
A finished file transcription (the interface is in Japanese)

Drop in a file and press Start. That's all.

Apple IntelligenceUses the AI built into your Mac

The AI tidies your words,
and nothing you say leaves your Mac.

Press ⌥End, talk, and press it again: the text is typed at the cursor of whatever app you are using, from mail and chat to notes and AI prompts. Speech recognition and the AI that tidies your words both run on the Mac, so you get dictation for free without signing up for a monthly cloud service.

  • Your voice never leaves the MacRecognition uses Apple's on-device speech recognition or Whisper. No recordings or text are saved to files.
  • No subscription, no accountTalk as long as you like without watching minutes or word counts.
  • AI styles for each purposeFillers are removed, and Apple Intelligence rewrites on the Mac for mail, chat, search or AI prompts.
  • A style per appShort and casual for chat, polite for mail. Switch with ⌥↑↓ while talking; the dictionary fixes names.

What you saidえーと、来週の定例で使う資料をまとめてほしいんですけど、あのー、グラフは3つまでとか、あと、です・ます調でお願いします

What was typed (AI prompt, outline)【依頼内容】 来週の定例で使う資料をまとめてください。 【制約条件】 ・グラフは3つまで ・です・ます調

Typical paid cloud dictation compared with Whisper Local dictation
Typical paid cloud dictationWhisper Local dictation
AI that tidies the textAI on the provider's serversApple Intelligence built into the Mac (on the Mac)
PriceOften a monthly or yearly planFree (no ads, no payments)
Where your voice goesProcessed on the provider's serversProcessed only on this Mac
Sign-upAccount requiredNone
InternetNeeded every timeNot needed after setup
UsagePlans may have limitsNo limits

Cloud services differ; the comparison above shows typical cases. The app and its styles are in Japanese. Dictation needs macOS 14 or later on Apple silicon; AI styles need macOS 26 or later with Apple Intelligence turned on. Typing into other apps needs the macOS Accessibility permission. Whisper models or Apple's recognition data may be downloaded once at first.

Before you download

What to check before you download

Whisper Local at a glance
Price
FreeNo account or API key needed
Requirements
macOS 14 or later, Apple silicon (M1 or later)Intel Macs are not supported. Models need about 280 MB to 1.8 GB of free space (150 MB–1.6 GB for the model plus working space)
App interface
JapaneseTranscription supports Japanese, English, auto-detect and many other languages
Permissions
Microphone, for live transcription and dictationmacOS may also ask for Speech Recognition when you use the built-in recognizer. Typing dictated text into other apps needs Accessibility permission (without it, the text is copied to the clipboard). Transcribing files with Whisper needs no permissions
Network
Hugging Face (model download) and this site (update check)The update check can be turned off. Audio and results are never sent
Where data is stored
Only on this Mac~/Library/Application Support/WhisperLocal
Latest version
0.3.28 (November 5, 2026)
Features

What it does

Whisper Local uses OpenAI Whisper models (Core ML versions) to transcribe on your Mac. Once a model is set up, transcription works offline.

Audio and video files

WAV, M4A, MP3, MP4, MOV and other formats macOS can read, up to 2 hours per file. Choose Whisper (accurate, multilingual) or Apple's built-in speech recognition (fast, macOS 26 or later) as the engine.

Live transcription

Pick a microphone and speak; text appears as you talk. Choose the built-in macOS engine (starts instantly) or Whisper (higher accuracy).

Dictation in any app

Press ⌥End and speak; the text goes in at the cursor of the app you are using. A “clean up” style removes filler words, and on macOS 26 or later with Apple Intelligence turned on, styles for email, chat and more rewrite the text on your Mac. You can set a style for each app, and even for each browser tab by a word in the window title.

Edit and export in 5 formats

TXT, CSV, SRT, VTT and SBV. Segments with reversed or overlapping subtitle times are flagged for review before export.

Speaker labels (optional)

Splits the transcript by speaker. Setting the number of speakers (2–8) reduces mix-ups. You can rename speakers and fix assignments; changes apply to every export format.

Dictionary and punctuation

Register common misrecognitions as "wrong → right" to replace them automatically. Japanese punctuation can be added to spoken text (on/off); it is not inserted after English or other non-Japanese words.

History and menu bar

Original results and your edits are saved on your Mac and restored next time. The app can stay in the menu bar and accept requests from apps you allow on the same Mac.

Privacy

Network and data

Whisper Local connects to the network in only these 2 cases. In neither case does it send your audio, transcripts, edits, history or dictionary. It sends no usage statistics.

WhenWhereWhatCan it be turned off?
Model downloadHugging FaceDownloads model files. No audio or results are sentOnly happens when you ask for a model
Update check (after launch and once a day)This site (k386.sub.jp)Fetches update information. The installer is downloaded only if you choose to updateYes, in Settings
  • Whisper transcription runs in a separate process that is blocked from network access.
  • The built-in macOS speech recognition (for files or live transcription) processes speech on the device. If recognition data for a language is missing, macOS may download it from Apple.
  • The microphone is used only between Start and Stop, and while dictation is on (from pressing the dictation key until you press it again or cancel).
  • Dictation audio and text are not saved to files (only the last 20 texts are kept as history, which you can turn off). Styles that use Apple Intelligence also run only on your Mac.
  • Other apps on the same Mac (such as icRec) can send transcription requests only from inside the Mac (127.0.0.1) and only for folders you allow. Connecting icRec takes one confirmation: press “Whisper Local と連携する” in icRec, then “連携する” (Connect) in Whisper Local; the setup is filled in automatically and no token is put in a URL.
FAQ

FAQ

Why is it free?

I built it for my own work and want other Mac users to be able to use it too. There are no ads and no paid version.

Does it work on Intel Macs?

No. It requires a Mac with Apple silicon (M1 or later) and macOS 14 or later.

Is it accurate?

Speech recognition and speaker labels can contain errors. Always review the results before using them.

What are the limits?

Files are processed one at a time, up to 2 hours each. Apple's built-in recognizer (macOS 26 or later) cannot auto-detect the language, so "Auto" uses your Mac's language. The first Whisper run (also the first run after an update) converts the model for your Mac and can take several minutes. Dictionary replacement does not consider context.

How do I uninstall it?

Delete "Whisper Local" from the Applications folder. Models, history and live recordings are stored in ~/Library/Application Support/WhisperLocal; delete that folder too if you no longer need them. Settings and the dictionary are in ~/Library/Preferences/local.duca.whisper.plist. If you changed the save folder for live recordings or texts, check that folder too.

Read more (in Japanese)

Longer articles about Whisper Local on note and Zenn. All articles

Download Whisper Local

Version 0.3.28, released November 5, 2026.

Size
5.1 MB
Requirements
macOS 14 or later, Apple silicon
Notarization
Notarized by Apple
Price
Free
Trademarks

Trademarks

  • Apple, Mac, macOS, Core ML and Numbers are trademarks or registered trademarks of Apple Inc. in the U.S. and other countries and regions.
  • OpenAI is a trademark or registered trademark of OpenAI. Whisper is the name of a speech recognition model released by OpenAI. Whisper Local is an app by K386 and is not provided or endorsed by OpenAI.
  • Hugging Face is a trademark of Hugging Face, Inc.
  • Excel is a trademark or registered trademark of Microsoft Corporation.
  • Python is a registered trademark of the Python Software Foundation.
  • Intel is a trademark of Intel Corporation or its subsidiaries.
  • Other company, product and service names on this site are trademarks or registered trademarks of their respective owners. ™ and ® marks are omitted in the text.
  • K386 and Whisper Local are not affiliated with, endorsed or sponsored by any of the companies above.