Download Whisper Local
For Macs with macOS 14 or later and Apple silicon (M1 or later). Intel Macs are not supported.
Whisper Local 0.3.28
Latest version, released November 5, 2026. A DMG file (5.1 MB).
By downloading, you agree to the Privacy Policy & Disclaimer.
How to verify the downloaded file
Run these two lines in the Mac’s Terminal app to check that the file is the official one (assuming the file is in your Downloads folder).
spctl -a -vv -t open --context context:primary-signature ~/Downloads/Whisper-Local-0.3.28.dmg shasum -a 256 ~/Downloads/Whisper-Local-0.3.28.dmg
- If the first line prints “accepted” and “source=Notarized Developer ID”, the file has been notarized by Apple.
- If the value from the second line matches the SHA-256 in the table above, the file is unchanged since it was published.
Notarization is Apple’s automated check for known malware. It does not guarantee the app’s quality or behavior.
How to install
If you use Whisper, a model is downloaded first (150 MB–1.6 GB), so start while connected to the internet.
App interface: Japanese. Japanese labels are shown in parentheses, for example Settings (設定).
Only the first Whisper transcription may take from a few minutes to about 10 minutes.

-
Open the DMG
Double-click the downloaded “Whisper-Local-0.3.28.dmg”.
-
Drag it to the Applications folder
Drag “Whisper Local” onto “Applications” next to it.
-
Open the app
Open “Whisper Local” in the Applications folder. The first time you open it, macOS may show a screen asking you to confirm that it is an app downloaded from the internet. If you downloaded it from this page, choose “Open”.
-
Prepare a model in the setup guide
The setup guide opens on first launch. Choose a language and press “Prepare this model” (このモデルを準備する); the model is downloaded and set up (on macOS 26 or later you can also choose “Mac built-in speech recognition” (Mac標準の音声認識), which needs no download). After choosing speaker labels and the export format, you can try it with “Try with sample audio” (サンプル音声で試す). The default Large v3 Turbo is about 1.6 GB. You can also choose the smaller Turbo Compact (about 650 MB) or the English-only Base English (about 150 MB). You can open the guide again at any time from “Setup Guide…” (セットアップガイド…) in the menu or “Open Guide” (ガイドを開く) on the “Model” (モデル) tab in Settings.
-
Drop in audio or video and start
Drop a file onto the window and press “Start Transcription” (文字起こしを開始) (⌘↩). The first time only, the model is converted for this Mac, which takes from a few minutes to about 10 minutes. Progress is shown on screen. After that it takes much less time.
The first time you use live transcription, you are asked for permission to use the microphone (when you use Mac built-in speech recognition, macOS may also ask for Speech Recognition permission). See the User Guide for how to use it.
If you use Homebrew, you can also install it by running the following in Terminal. It downloads the same file as the official site and checks the SHA-256 before installing.
brew install --cask kenji-miygauchi/apps/whisper-local
Requirements
| macOS | macOS 14 Sonoma or later |
|---|---|
| Mac | Apple silicon (M1 or later). Intel Macs are not supported |
| Free space | About 12 MB for the app + 150 MB–1.6 GB for the model (depending on the model you choose; about 11 MB more if you use speaker labels) + working space (while downloading, you need the size of the model plus about 130 MB free) |
| Memory | About 1.5–2 GB while processing with the default model (varies with the audio length and the model) |
| Mac built-in speech recognition | macOS 26 or later, both as the file transcription engine and as the live engine (earlier versions of macOS use Whisper) |
| Live transcription | A microphone is needed |
| Network | Connects to Hugging Face when downloading a model, and to this site to check for and download updates. When you use Mac built-in speech recognition, macOS may download language data from Apple. Audio and transcription results are not sent. Transcription itself works offline |
| Price | Free. No account or API key needed |
| App interface | Japanese |
How to update
About 5 seconds after launch, and then once a day, the app checks this site for a new version. If one is found, it is shown at the top of the window (and also in the menu bar if the app stays in the menu bar). If you choose “Update and Restart” (アップデートして再起動), the app checks the file size and SHA-256, Apple’s notarization and the distributor’s signature before replacing itself. Models, history, edits and settings are kept.
If you are running the app directly from the DMG, or it is in a location that cannot be written to, it cannot replace itself automatically, so “Open Download Page” (ダウンロードページを開く) is shown. Download it from this page.
You can turn off the automatic check with “Check for updates automatically” (アップデートを自動で確認する) in Settings. Even when it is off, you can check with “Check Now” (今すぐ確認) in Settings or “Check for Updates…” (アップデートを確認…) in the menu, and update inside the app. To replace it manually, download the new DMG from this page and replace the app in the Applications folder following the same steps as in “How to install” above.
How to uninstall
- If the app is in the menu bar, quit it.
- Move “Whisper Local” from the Applications folder to the Trash.
- Downloaded models, history, and live recordings and texts are in
~/Library/Application Support/WhisperLocal. Delete this folder too if you no longer need them. - If you changed the save folder for recordings or texts, check that folder too.
- The app’s settings are stored in
~/Library/Preferences/local.duca.whisper.plist.
If you turned on “Launch at login” (ログイン時に自動で起動する), check that it is no longer listed under General → Login Items in System Settings. Credentials for connected apps are stored in the keychain under the name “local.duca.whisper.api”. If you no longer need them, delete them in Keychain Access, or remove the registrations under “Integration” (連携) in Settings before deleting the app.
Release notes
0.3.28 (build 31) — November 5, 2026
- In the dictation “Per-app settings” (アプリごとの設定), you can now specify a word contained in the window title. Even in a browser, you can separate tabs, for example AI prompt for a ChatGPT tab and Mail for a Gmail tab. Which setting is used is decided by the app and window in front when you finish speaking.
- In styles that use AI, when the AI result cannot be used, the app now asks the AI once more. When only part of a long text cannot be used, only that part is replaced with the “Tidy” (整える) text.
- “Custom” (カスタム) now lets rewrites such as bullet lists and summaries through more easily. The instruction field shows a guide to the instruction length that the built-in AI follows more easily, and the default instruction was made into an example.
- When dictated text is typed in without problems, the band at the bottom of the screen now disappears right away. Previously, it sometimes did not disappear while the mouse was over the band.
- When there is no transcription result, the main window now shrinks to the height of its contents.
0.3.27 (build 30) — November 3, 2026
- File transcription now skips parts without human voices (breathing in sleep, noises, long silences). Previously, sentences that were never actually spoken, such as “ご視聴ありがとうございました” (Thank you for watching) or “おやすみなさい” (Good night), were sometimes written into such parts.
- Quiet narration over music (such as a video playing nearby) may not be detected as voice and may not be transcribed.
0.3.26 (build 29) — October 22, 2026
- Added “Clear” (クリア) to live transcription. After stopping, you can clear the displayed text. Saved texts and recordings are kept.
0.3.25 (build 28) — October 22, 2026
- With the “▼” to the right of the save button in the result window, you can now change the save format for the displayed result only (the default in Settings does not change, and the next result goes back to the default).
- Added “Clear” (クリア) to the result window. It clears the displayed result and returns to the first screen. The result is kept in the history.
- TXT and CSV can now be saved without the time check even when the split is “By seconds” (秒数ごと). The time check is only for saving subtitles (SBV, SRT, VTT).
- The dictation “AI prompt” (AIへの指示) style now keeps sentences that check something or make a guess as they are.
- When “AI prompt (outline)” (AIへの指示(項目立て)) cannot tidy the text, it now tries again with “AI prompt” before falling back to “Tidy”.
0.3.24 (build 27) — October 11, 2026
- Added the dictation style “AI prompt (outline)” (AIへの指示(項目立て)). It organizes the prompt you send to an AI into “【依頼内容】” (Request) and “【制約条件】” (Constraints), listing the conditions you spoke with “・”. If you did not speak any conditions, only 【依頼内容】 is typed.
- “AI prompt” now turns spoken expressions such as “〜とか〜とか” and “あと” (and also) into written language.
- Fixed an issue where the AI check sometimes mistook a polite request such as “〜ていただけますでしょうか” (could you please…) for a question and skipped it.
- When you have rewritten the instruction for a style that uses AI, Settings now shows “Using a rewritten instruction” (書き換えた指示を使用中) and the style list shows “(rewritten)” ((書き換え)). When an update brings a new default instruction, Settings tells you, and you can switch with “Use the new default” (新しい初期値にする). The instruction field is saved only when you press “Save” (保存).
- Under “Default style” (既定のスタイル) in Settings, the apps whose per-app style takes priority are now shown.
0.3.23 (build 26) — October 11, 2026
- Revised the dictation style “AI prompt”. It now tidies the text without leaving out background, reasons or conditions, and without turning explanatory sentences into requests. If you have rewritten the instruction in Settings, press “Reset to Default” (初期値に戻す) to use the new instruction.
- When the AI result leaves out something you said, or turns an explanatory sentence into a request, the “Tidy” text is now typed instead of the AI result. When the AI starts writing sentences you did not say, it is stopped early.
- The dictation band now closes right away when the text goes in successfully. When there is something to tell you, it is shown for 8 seconds and does not close while the mouse is over it. When the engine is Whisper, the band shows that finishing takes a few seconds.
- Dictionary replacement now works even with variations in katakana spelling (with or without “ー”, or small “ィ” versus “イ”) (for registered words of 4 or more characters).
0.3.22 (build 25) — October 11, 2026
- You can now set the dictation style and input method for each app (“Dictation” (音声入力) → “Per-app settings” (アプリごとの設定) in Settings). For example, if you set “Mail” (メール) for your mail app and “Chat” (チャット) for your chat app, you can use them without switching.
- You can now switch styles while speaking with ⌥↓ (next) and ⌥↑ (previous).
0.3.21 (build 24) — October 11, 2026
- Added dictation styles that tidy text with the Mac’s built-in AI: “AI prompt”, “Mail”, “Chat”, “Search” (検索) and “Custom” (on macOS 26 or later, when Apple Intelligence is turned on). Processing is done only on this Mac.
- When the app judges that the AI result strays from what you said (it answered a question, added something you did not say, and so on), it types the “Tidy” text instead of the AI result and shows the reason. The AI instruction for each style can be rewritten under “Dictation” in Settings.
- The dictation band now has a dark color scheme and shows a waveform that moves with your voice, so you can check that the microphone is picking up your voice. You can switch styles from the band’s pop-up menu, even while speaking.
- The Accessibility permission status now updates automatically when you turn it on in System Settings.
0.3.20 (build 23) — October 11, 2026
- Added “Dictation” (音声入力), which lets you type by speaking whatever app you are using. Press ⌥End (⌥fn→ on keyboards without an End key), speak, and press it again; the text is typed at the cursor. While you speak, a small band appears at the bottom of the screen. The key can be changed under “Dictation” in Settings.
- You can choose the style of the typed text from “As is” (そのまま), “Record” (記録) and “Tidy” (整える; removes fillers such as えーと and あのー). Dictionary replacement works with every style.
- To type at the cursor, Whisper Local needs permission under “Accessibility” in System Settings. Without the permission, the text is copied to the clipboard. Dictation steps were also added to the setup guide.
0.3.19 (build 22) — October 2, 2026
- You can now choose Mac built-in speech recognition (macOS 26 or later) under “Model” in the setup guide. You can try it and start using it right away without downloading a Whisper model.
- The update description and the confirmation message shown when saving subtitles now use the same wording as the actual screens.
0.3.18 (build 21) — October 2, 2026
- You can now choose the engine for file transcription. Under “Engine” (変換エンジン) in “Transcription” (文字起こし) on the “Model” tab in Settings, you can switch between “Whisper (accurate, multilingual)” (Whisper(高精度・多言語)) and “Mac built-in speech recognition (fast)” (Mac標準の音声認識(高速)). Mac built-in speech recognition (macOS 26 or later) needs no model download or first-time conversion. Speaker labels, segment review, subtitles and the dictionary work with either engine.
- Segments where Whisper repeats the same words are now reprocessed automatically.
- “、” and “。” are no longer inserted after English or other non-Japanese words.
- The language being transcribed is now shown during processing. With “Auto-detect” (自動検出), the language detected most often across segments becomes the language of the result.
0.3.17 (build 20) — September 29, 2026
- Added “About This App” (このアプリについて). From “About This App” in the “Whisper Local” menu (or the menu bar icon when the app stays in the menu bar), you can check the version, the copyright notice (© 2026 K386) and a link to the official site. It is shown once on first launch and on the first launch after updating to a new version.
0.3.16 (build 19) — September 25, 2026
- Shortened the wait for speaker labels. When speaker labels are on, the speaker model is prepared in the background when the app launches and when you turn the option on, so the first transcription no longer waits a few seconds (the same applies right after an update).
- Added “Number of speakers” (人数) to “Speakers” (話者) in Settings. If you know how many people are in the conversation, specifying 2–8 reduces speaker mix-ups. If you don’t know, you can leave it at “Estimate automatically” (自動で推定).
0.3.15 (build 18) — September 25, 2026
- Connected apps such as icRec can now be connected with a single confirmation. When you press “Whisper Local と連携する” (Connect to Whisper Local) in the connected app, a confirmation appears in Whisper Local; just press “連携する” (Connect), and staying in the menu bar, folder permission, registration of the connected app and accepting requests are all set up automatically. There is no need to copy IDs or tokens by hand.
- When launched from a connected app such as icRec (--background), the app now also starts in the menu bar without opening a window if “Stay in the menu bar when the window is closed” (ウインドウを閉じても常駐する) is on.
0.3.14 (build 17) — September 24, 2026
- Fixed a bug where, when launched automatically at login to the Mac, the window sometimes opened even though “Stay in the menu bar when the window is closed” was on. When launched automatically right after login, the app now starts in the menu bar without opening a window, and a banner at the top right of the screen tells you for a few seconds (click the banner to open the window).