WhisperDrop 2.6.1

WhisperDrop 2.6.1

This update makes notes safer to leave open and fixes microphone startup and recording recovery issues.

Notes

• A recording pauses when your Mac sleeps or its lid closes. It stays paused until you choose Resume, and the note explains why it paused.
• A microphone that stops unexpectedly pauses the note and shows a warning. Dictation attempts to finish the captured speech and warns that it may be incomplete.
• Stop a note while microphone access is still pending without waiting for the permission prompt to finish.
• Audio and transcript folders, and kept recordings, use the note's title. Renaming a note also renames its managed files.
• Kept recordings use AAC M4A at 48 kbit/s. Older managed CAF recordings are converted after verification, with the originals moved to the Trash. If an M4A already exists at the destination, both files are kept and the app reports the problem.
• Preserve pending library renames if a conflicting file prevents recovery. A recording or conversion warning remains visible even when no speech was detected.

Microphone and dictation

• Fix a startup crash after an input device changes to a different sample rate. The recording tap uses the current hardware format.
• Report microphone setup errors instead of closing the app. The fix applies to dictation and microphone notes.
• Settings, Dictation shows an Allow action when Input Monitoring is missing. Shortcut monitoring retries when the app becomes active after a permission change.

If the shortcut does nothing, allow WhisperDrop 2 in System Settings > Privacy & Security > Input Monitoring. Allow Accessibility to insert text into other apps and Microphone to record your voice. macOS can request these permissions again after an update. If monitoring still does not start, quit and reopen the app.

Install

Apple Silicon, macOS 14 or later. Download the DMG attached to this release and replace the app in Applications. Notes, settings and downloaded models stay outside the app bundle.

This release is available by manual download only. The signing key for the existing Sparkle update feed is unavailable, so Check for Updates does not offer 2.6.1. The existing 2.6.0 feed remains unchanged.

The app is locally signed and not notarized. A first downloaded copy may require System Settings > Privacy & Security > Open Anyway.

Validation

The debug and release suites passed 167 tests each, with 11 optional checks skipped. An isolated native build was exercised with microphone capture, simulated sleep, manual resume, note saving and renaming, and a synthetic Italian file transcribed with Parakeet. Two-minute synthetic note replays at normal speed and at 8x reported no dropped packets. These are controlled tests, not a guarantee of accuracy or lossless capture. Physical lid closure, actual system sleep, a Bluetooth device switch and cross-app dictation still need verification.

WhisperDrop 2.6.0

WhisperDrop 2.6.0 makes Parakeet v3 the default speech model, writes notes while you speak with one sentence per paragraph, lets you pause a note and rename it by clicking its title, and adds local writing tools and in-app updates.

Install

Apple Silicon, macOS 14 or later. Coming from 2.5.0 or earlier, download the DMG once and replace the app; your notes, models and settings stay. From 2.6.0 on, use WhisperDrop 2 > Check for Updates… or Settings > General > App updates instead of downloading again.

The app is ad hoc signed and not notarized, so macOS may block the first launch. If you trust this build, open it once, then choose System Settings > Privacy & Security > Open Anyway. After replacing the app, macOS may ask again for Accessibility and Input Monitoring.

Parakeet v3 by default

• Parakeet v3 is the default model and is no longer experimental. It covers 25 European languages, including Italian and English, and is fast enough to write while you speak.
• If your main model was Turbo and Parakeet is already downloaded, the app switches to Parakeet once. Choosing another model afterwards is kept.
• Whisper models are listed as legacy models in Settings > Models. When your language is not one of Parakeet's 25, an installed Whisper model is used automatically, Turbo first, and Settings shows which one.

Notes

• With Parakeet, notes are no longer cut into one-minute chunks. Words join the transcript a few seconds after you say them, and every sentence starts a new paragraph. Words that may still change are shown in grey.
• Pause stops recording without closing the note. Paused time is not recorded and the timer stops; Resume continues in the same note and audio files. While paused, the model can unload to free memory.
• Click a note's title to rename it, while recording or after saving. Press Return to save or Escape to cancel.
• With a legacy Whisper model, notes still appear after a pause or about a minute of speech.

Writing tools

• Writing tools open in the main window. Fix grammar, make wording natural, adjust tone, rewrite, summarize, organize notes or make text concise. Edit the presets or add your own in Settings > Writing.
• Select text in another app and press Control + Option + D for a compact review panel. Replace is offered only when the original selection can be verified.
• Completed notes and transcripts have a Writing menu. Summaries and revisions are saved as separate versions; the original transcript stays.
• A review panel can open after dictation. Automatic note summaries are optional and off by default.
• Text models run locally: LFM2.5 2.6B (default, 1.59 GB) or MiniCPM5 1B (experimental, 688 MB), chosen in Settings > Models > Text editing. Review every suggestion; both models can make mistakes, especially in Italian.

In-app updates

• Updates are signed and verified before installation, and install only after you confirm. Daily checks are optional and off by default.
• Updating waits for recordings, dictation, transcriptions, imports and writing actions to finish. Models, notes and settings are kept.

Details

• Lecture transcription and streaming notes
• Writing tools
• In-app updates

WhisperDrop 2.5.0

WhisperDrop 2.5.0 has one place to choose the model, live text that keeps everything you said, and dictation that can write into the text field while you speak. It also includes the experimental Parakeet v3 engine and saved-note renaming, prepared for 2.4.0 but not published separately. Whisper Turbo Q5 remains the default.

Changes since 2.3.0

One place for models

• Settings > Models is the only place to download, delete and choose a model. Before, the main window chose the model for files and Settings chose it for dictation and notes, and the two could differ without warning. The main window and the sidebar now show the model and open Settings > Models.
• The model you had chosen in Settings becomes the model for everything.
• Advanced: a different model for each task can give dictation, notes or files their own model, for example Parakeet v3 for dictation and Whisper for lectures.
• Keep model ready in the menu bar and the Unload model setting were two controls for the same thing. Keep model ready is now one of the Unload model choices; turning it off in the menu bar restores your previous interval. Keep audio has one name beside New note and in Settings.
• Notes and dictation check that their model is downloaded before the microphone opens. Before, a note could record with a missing model and skip every part of the transcript.

Parakeet v3, experimental

• Download Parakeet TDT 0.6B v3 with a 4-bit encoder, about 489 MB. It runs locally in a native Swift helper using MLX and supports 25 European languages, detected automatically. It ignores language hints and vocabulary lists. Unsupported language selections are refused before capture starts.
• Parakeet file imports use pause-aligned windows of up to 35 seconds, in a process separate from the dictation and note model.

Live text, with Parakeet

• Notes show every word heard since the last confirmed paragraph, in grey. Before, the grey text covered only the last 20 seconds, so earlier unconfirmed words disappeared during long passages. The transcript follows new text only while you are at the bottom; scroll up to read earlier text and use Latest to return.
• Dictation can write into the text field while you speak (Settings > Dictation > Live text > In the text field). Native text fields show every word and correct it in place; when you finish, the whole dictation is decoded once more and the field is corrected to that text. Browsers and apps whose fields ignore accessibility edits receive words once they settle, about six seconds behind, and the newest words show in the dictation bar; there the typed words stay and only the rest is decoded at the end. Press Escape to discard the dictation, including the words already written. If you click elsewhere or type while dictating, WhisperDrop stops writing and copies the full text to the clipboard instead.
• The dictation bar shows up to four lines of live text instead of one truncated line.
• Live requests stay short however long you speak: words settle once about six seconds of audio follow them, and later requests start ten seconds before the settled words, so each request covers at most 30 seconds.

Other changes

• Dictation that starts right after other text, such as "end." with the caret after it, begins with a space instead of joining the previous word.
• Rename a saved note with Rename… beside its title or from its library context menu.
• Whisper dictation keeps the full encoder window. The shortened window of 2.3.0 returned unrelated words and duplicated sentences on short clips.

Validation and limits

• 114 standard tests pass in debug and release builds; seven model replays that need local recordings are skipped by the standard suite and were run separately.
• Live text was replayed silently through the real model host on Italian and English recordings cut into dictations of about one minute. Against decoding each dictation whole, word error rate went from 7.22% to 7.36% on Italian audiobooks, 4.91% to 5.11% on Italian TED talks and 1.41% to 1.28% on English audiobooks. The wait after a dictation ended dropped from about 1.7 to 2.2 seconds to about 0.6 seconds when only the unsettled end was decoded. Details are in docs/parakeet-evaluation.md.
• A 4-minute note replayed at recording pace kept every sample; its grey text grew to 121 words, 56 seconds of speech not yet confirmed.
• Dictation into the text field was checked with recorded clips in TextEdit, the Safari address bar, Safari and Chrome web text areas, after existing text, with the caret moved mid-dictation and with Escape. It was not tested with a live microphone, a physically held key, Electron apps (Slack, Notion, VS Code), Microsoft Word or terminals. Password fields turn live writing off.
• In browsers, words appear about six seconds after you say them, and a slow request can make them later.
• The packaged app transcribed an Italian clip with Parakeet v3 and with Whisper Turbo Q5.
• Tested only on macOS 26 on Apple Silicon.

Install

Apple Silicon, macOS 14 or later. Quit the previous version and replace the app from the DMG. Models, notes and settings are stored separately from the application bundle.

This build has an ad hoc signature and is not notarized. Gatekeeper may block a downloaded copy. If you choose to trust the build, use Apple's Open Anyway flow in System Settings, Privacy & Security. Because the signature changes with every ad hoc build, macOS may ask again for Accessibility, Input Monitoring and Microphone access after updating; if dictation does nothing, remove WhisperDrop 2 from those lists in Privacy & Security and allow it again.

The app grows from about 144 MB in 2.3.0 to about 477 MB, because it includes Parakeet and two MLX shader libraries. It requires no Python or cloud service at runtime.

Release assets include SHA256SUMS.txt and the unchanged FFmpeg 8.1.2 source archive. Runtime license notices are bundled in the app.

WhisperDrop 2.3.0

A short guide now opens on first launch, the app uses one spoken language everywhere, and dictated text is cleaner.

Changes

• New six-step guide: dictation, notes, files and YouTube, model and language, then permissions. It appears once, for new installs and for people updating from earlier versions. Reopen it from Help, WhisperDrop 2 Guide, or Settings, Permissions.
• One spoken language for files, dictation and notes. Change it in the main window, the menu bar or Settings, General. Your earlier choice is kept. Note language, beside New note, now applies to the next note only.
• Dictation no longer capitalizes words in the middle of a sentence after a pause. Short dictation is decoded as one passage, and longer dictation fixes capitals where passages join. Add names to Vocabulary to keep their capitals.
• Dictation that repeats the same phrase back to back is transcribed once more with the full audio window. The new text is used only when it repeats less and keeps the rest of what you said.
• The About panel no longer lists the original project. The README and NOTICE still credit it.

Turbo Q5 remains the default. Transcription runs locally. Dictation, notes, file imports, YouTube videos and playlists remain available.

Validation and limits

82 automated tests passed in debug and release. Four optional tests that need local models or fixtures were skipped in the standard run. Three synthetic Italian clips of 21, 23 and 34 seconds went through the app's dictation path with Turbo Q5. The two short clips came back as one passage each, with correct punctuation and capitals, and one word was spelled correctly that the previous settings got wrong. The long clip joined cleanly. Latency matched the previous settings. These synthetic clips do not measure accuracy on real speech. The loop retry was tested with sample text only, because a real loop could not be reproduced on demand. In dictation longer than 28 seconds, a name said only once right after a pause can lose its capital unless it is in Vocabulary.

The bundled runtime tools are identical to those in 2.2.0.

Install

Apple Silicon, macOS 14 or later. Download the DMG, quit the old version, and copy WhisperDrop 2.app into Applications, replacing the previous copy. Models, notes and settings survive replacement.

This release has a local code signature and is not notarized. Gatekeeper may block the first launch after a browser download. If you choose to trust this build, try opening it once and follow Apple's Open Anyway instructions in System Settings, Privacy & Security.

SHA256SUMS.txt contains release-asset checksums. FFmpeg-8.1.2-source.tar.xz contains the unmodified FFmpeg source used for this build; its configuration is in scripts/prepare-runtime.sh. Runtime license notices are bundled in the app and documented in the repository.

WhisperDrop 2.2.0

Recordings and transcripts now default to Documents/WhisperDrop, with separate Audio and Transcripts folders. This release also adds selective quiet-speech gain and checks for invented closing phrases and sustained repetition.

Changes

• Open Settings directly from the main window.
• Choose a saved-files destination in Settings, General, Saved files. Existing managed files move with the library. Copies are verified before the originals are removed, and interrupted cleanup resumes after relaunch.
• Migrate earlier saved recordings and transcripts from Application Support automatically. Notes retain their original audio and export TXT, SRT and VTT transcripts.
• Enable Boost quiet audio by default for notes and imported files. Gain applies only when speech is quiet and sufficiently above the noise floor, with peak protection and a local voice detector. Original recordings and imported files stay unchanged.
• Check short Italian and English closing phrases against voice activity. Spoken phrases such as "Grazie" and "Ciao" are retained; unsupported closings in silence are removed.
• Retry sustained transcription loops once on original audio without previous text context. A candidate replaces the first result only when its repetition score improves. Unresolved passages keep a visible warning.

Turbo Q5 remains the default. Dictation, file imports, YouTube videos and playlists remain available. Transcription runs locally. Denoising is not enabled because it did not consistently improve the tested recordings.

Validation and limits

70 automated tests passed in both debug and release. Three optional fixture-dependent tests skip in standard runs and were exercised separately. Eight real-model controls checked quiet spoken closings, silence, room noise and repeated sentences. A normal-cadence recording replay preserved all input samples within PCM16 quantization; a separate replay verified deletion when Keep audio is off.

The complete September 30 lecture replay reduced isolated closing phrases from ten to zero. Gain was skipped because the background noise was too high. Recognition errors and shorter repetitions remain, and there is no independent reference transcript for a lecture accuracy score. An accelerated 10x capture stress test dropped one packet and reported the loss.

The release app, disk image and nested signatures were checked. See test methods and limits.

Install

Apple Silicon, macOS 14 or later. Download the DMG, quit the old version, and copy WhisperDrop 2.app into Applications, replacing the previous copy. Models, notes and settings survive replacement. On first launch, earlier saved files move to Documents/WhisperDrop.

This release has a local code signature and is not notarized. Gatekeeper may block the first launch after a browser download. If you choose to trust this build, try opening it once and follow Apple's Open Anyway instructions in System Settings, Privacy & Security.

SHA256SUMS.txt contains release-asset checksums. FFmpeg-8.1.2-source.tar.xz contains the unmodified FFmpeg source used for this build; its configuration is in scripts/prepare-runtime.sh. Runtime license notices are bundled in the app and documented in the repository.

WhisperDrop 2.1.1

Lecture notes now wait for natural pauses, with a 60-second maximum and audio overlap at forced cuts. The update fixes audio packet ordering and removes the server wrapping that split words into separate lines.

Changes

• Preserve two seconds of audio across forced cuts and complete boundary words on the next request.
• Group the reading view and copied text into paragraphs, keeping subtitle timing separate.
• Show note language and Keep audio beside New note, with a first-use language reminder.
• Reveal retained recordings in Finder and offer optional subject vocabulary in Settings.
• Move disk writes off the capture worker and preserve capture, save and transcription warnings in completed notes.
• Keep legitimate repeated sentences and exclude finishing time from recording duration.

Turbo Q5 remains the default. Dictation, file imports, YouTube videos and playlists remain available. Transcription runs locally.

Validation

49 automated tests passed. The two opt-in inference tests skip during the normal suite and were run separately, including silent real-time capture replay, retention-off replay and a complete 58-minute lecture replay. The release app, disk image and nested signatures were checked. See test methods and recognition limits.

Install

Apple Silicon, macOS 14 or later. Download the DMG, quit the old version, and copy WhisperDrop 2.app into Applications, replacing the previous copy. Models, notes and settings are stored outside the app and survive replacement.

This release has a local code signature and is not notarized. Gatekeeper may show "Apple cannot check the app for malicious software" after a browser download. That is a verification warning; Apple uses a different alert for detected malware. If you choose to trust this build, try opening it once and follow Apple's Open Anyway instructions in System Settings, Privacy & Security. A local signature cannot guarantee warning-free installation on other Macs.

SHA256SUMS.txt contains the disk-image checksum. Runtime license notices are in the app and repository.

WhisperDrop 2.1.0

Local macOS transcription, with dictation and meeting notes. Apple Silicon, macOS 14 or later.

In this version

• Push-to-talk dictation into the app in front. The default key is fn.
• Notes from the microphone, system audio, or both. Transcripts label You and Others.
• A resident speech model that unloads after it has been idle. The menu bar can keep it ready.
• Settings, a menu bar controller, and a first-launch permission setup.

Turbo is the default model. It downloads inside the app. Audio stays on this Mac.

Install

Open the disk image and copy WhisperDrop 2 to Applications.

This copy is signed with a local development certificate and is not notarized. Gatekeeper can block it after download. Do not turn Gatekeeper off. If macOS refuses the app, right-click it, choose Open, then Open again. If macOS says the app is damaged, clear the download quarantine flag after copying it out of the image:

xattr -dr com.apple.quarantine "/Applications/WhisperDrop 2.app"

Before dictating, set System Settings, Keyboard, "Press 🌐 key to" to Do Nothing. Otherwise macOS also opens the emoji picker while you hold fn.

Source and third-party licenses are in this repository. The app also carries the runtime license files in Contents/Resources/Runtime.