New
Improved
Fixed
- ShoutFlow can now press Return for you after a dictation is inserted. It is off by default and gets enabled per app in the Advanced settings, so it only fires where you allowed it, for example in a chat app or the terminal where Return sends the message
- You can now cancel a dictation while it is being transcribed or enhanced, not just while recording. The X button in the overlay stops the whole process, and nothing is inserted
- Local Enhance starts faster: the on-device model now loads while you are still speaking instead of afterwards, so enhanced dictations are ready sooner
- Enhance handles numbers better and writes them as digits ("15" instead of "fifteen"), while keeping articles, expressions like "a couple of things", and years in their natural form
- Local Enhance is dramatically faster — enhancement with the on-device model now completes in about a second once the model is warm, instead of 15+ seconds
- Enhance is now significantly faster and more reliable — the local AI model stays warm in memory, so enhancement starts instantly instead of reloading the model for every dictation
- If Enhance can't complete (for example after a timeout), the app now inserts the plain transcription and briefly tells you — instead of failing silently
- Enhance could occasionally insert AI meta-commentary instead of your text — the output is now validated and the plain transcription is used whenever the result doesn't look right
- Parakeet model download no longer appears stuck at 2% — progress now updates continuously, and interrupted downloads resume where they left off
- After updating, existing installations reuse the already downloaded Parakeet model instead of re-downloading ~480 MB
- Transcription could silently stop working if the selected model wasn't downloaded — the app now falls back to an available model automatically
- The status overlay now shows the actual error message instead of just an icon
- Various small optimizations and under-the-hood improvements
- The last transcription preview in the menu bar now truncates to two lines so long text no longer overflows into other menu items
- New installs now default to English transcription instead of German
- Push-to-talk could stop recording too early when a second shortcut was configured and another modifier key (Shift, Control, Option) was tapped during the hold
- Setting both shortcuts to overlapping modifier combinations (e.g. Right Command alongside a ⌘-based custom shortcut) could trigger the wrong slot — the settings UI now prevents conflicting combinations
- Default Enhance CLI templates tuned for speed: Claude Code uses Haiku 4.5, Gemini CLI uses Gemini 2.5 Flash — both much faster than the previous defaults without sacrificing quality for transcript clean-up
- Enhance now runs optionally through a local CLI — Claude Code, Codex or Gemini CLI — much faster than the on-device LLM if you already have one installed
- Optional second keyboard shortcut dedicated to "transcribe + Enhance" — leave the first shortcut for plain transcription and trigger Enhance on demand by pressing the second one
- Lifetime transcription statistics no longer reset when you delete history items or hit the history size limit — stats are now tracked independently of the history
- Menu bar status overlay could appear duplicated or get stuck on screen, especially after connecting an external monitor
- Downloaded Whisper models could not be deleted after switching the transcription engine to Parakeet
- Parakeet is now the recommended model and auto-downloads during first setup
- Clearer license screen with a prominent "Get a License" button for new users
- Changelog accessible directly from Settings
- Parakeet is now significantly faster and more accurate
- Clearer error messages with visual distinction between warnings and errors, plus actionable hints
- Short or silent recordings now disappear quietly instead of showing an error
- Removed phantom filler phrases that Whisper occasionally hallucinated
- Transcripts are now clean, continuous text without stray line breaks
- Menu bar status widget redesigned — fixed size, rounded corners and a smooth slide-in animation
- Status widget no longer hidden behind the notch on MacBook Pro 14" and 16"
- Fixed a rare crash during Enhance
- Enhance — on-device AI that cleans up your transcripts, fixes grammar and handles self-corrections. Powered by Gemma 4, runs 100% locally
- History search and filter — find past transcriptions by text, date range or sort order
- Audio feedback — optional start, done and error sounds
- Audio ducking — automatically lowers system volume while you record
- Quick copy — double-click any transcript to copy it
- Floating icon remembers its position across restarts, even on multi-monitor setups
- License no longer lost after Homebrew updates
- Floating icon no longer gets lost when disconnecting an external monitor
- Statistics — see total transcriptions, duration, word count and language breakdown
- Left Command key as recording shortcut
- fn (Globe) key as recording shortcut
- File Transcription — drag and drop audio or video files onto the floating icon for a full transcript
- Right Command key as recording shortcut
- Real-time progress indicator for longer files, with cancel support
- Custom Words — add names and terms that are often misrecognized, WhisperBar corrects them automatically
- Clipboard Manager — choose whether to modify or preserve your clipboard when inserting text
- Right-click context menu on the floating icon — quickly change language, toggle push-to-talk or switch microphone
- Text is pasted significantly faster after dictation
- Text insertion more reliable across all apps
- Push-to-talk mode — hold the shortcut while speaking, release to stop
- Menu bar status indicator when the floating icon is hidden
- Private, fully offline speech-to-text powered by Whisper — right in your macOS menu bar
- 6 Whisper models from Tiny to Large v3 — choose speed or accuracy
- 10 languages supported
- Smart text insertion directly into the focused text field via Accessibility API
- Snippet expansion — trigger predefined text blocks with short voice commands
- Automatic language detection
- Transcription history
- Customizable keyboard shortcut
WhisperBar