Fix a Presspeech problem

The menu bar item is missing

macOS can hide third-party status items when a crowded or notched menu bar runs out of space. Open /Applications/Presspeech.app again; the existing process opens Setup Checklist. Enable Show in Dock there for a persistent alternative. To hide the item deliberately, choose Settings → Behavior → Show Presspeech in Menu Bar; Presspeech enables Dock access if needed, and the Dock menu can restore the item. Right-click the Dock icon for dictation controls, Settings, Support, and Quit.

A permission is missing or will not appear

On macOS 27 and later, System Settings calls Accessibility Device Control and Data Access, and Presspeech uses that current name in Setup Checklist. Open Support → Setup Checklist… and choose the affected Microphone, Accessibility / Device Control and Data Access, or Input Monitoring row. If the grant still does not appear, quit and reopen the copy in Applications, return to Setup Checklist, and choose Try Again. Presspeech then resets only the TCC service or services represented by that row and asks macOS again.

In the 0.3.8 download: an Accessibility row that appears granted does not separately verify permission to post the paste shortcut. Do not treat that row as proof that text reached the target; inspect the field and follow the paste recovery steps if delivery fails. From 0.3.9 (not in 0.3.8): the row checks both focused-window access and keyboard-event posting, and Copy Diagnostics reports those checks separately. Check GitHub Releases before relying on that change.

The speech model fails to load

Before reopening, retrying, or resetting macOS 0.3.8 with a missing or damaged model: another download may include a Hugging Face token inherited by Presspeech. The public model needs no account token. If one may be present in the environment that launches Presspeech—or you are unsure—do not start another download; when you can choose the timing, install macOS 0.3.9 before another download. Leave a working model cache in place. Read the guidance for existing 0.3.8 users before resetting anything.

If you choose to make another model request after reading the warning above, check the connection before retrying. If Presspeech reports an incomplete or corrupt cache and you still choose to download again, use Support → Reset Speech Model Cache…. This deletes the local model cache, then downloads and verifies a fresh copy; it does not remove settings or dictionary rules. From macOS 0.3.10 it resets only the selected model, so with Parakeet Ultra selected it downloads about 630 MB. Do not reset a working cache.

The hotkey does nothing

Confirm Settings → Hotkey shows the intended key and that Input Monitoring is granted in Setup Checklist. Use Settings → Hotkey → Reset Hotkey to Default to test Right Option. If a different app reserves that key, record another right-side modifier, F-key, or combination with Command, Control or Option. A custom combination consumes that shortcut globally, and conflicts cannot all be detected. For a custom combination, press exactly the saved modifiers; extra Command, Control, Option or Shift keys do not match.

Try Dictation works, but text is not inserted

Confirm Accessibility (Device Control and Data Access on macOS 27 and later) is granted in Setup Checklist, click the destination text field, then hold the hotkey without changing apps until transcription finishes. In the 0.3.8 download, Setup does not separately check keyboard-event posting, so an Accessibility row that appears granted does not prove the paste shortcut was accepted. If the row itself says Missing, use Try Again. From 0.3.9 (not in 0.3.8), Presspeech keeps the row Missing if either focused-window access or keyboard-event posting is unavailable, even when System Settings shows Presspeech enabled; its diagnostics identify both checks. If the HUD or menu says Copied — press ⌘V to paste in published 0.3.8, Presspeech copied the finished transcript when dictation completed. In builds with a separate starting-window warning, Can’t verify window — use ⌘V means it could not identify the exact window when recording began; the transcript was likewise copied. Builds with the revised notice say ⌘V if clipboard unchanged. These notices do not track later clipboard changes. If the clipboard still holds the transcript, return to the intended field and paste manually instead of dictating again. If the clipboard changed, use Copy Last Transcript when Recent Transcripts is enabled; otherwise the earlier copy may no longer be recoverable in Presspeech.

Presspeech also uses this safe fallback when it cannot verify the same destination. Some Electron/Chromium-based apps do not consistently expose the focused-window information the current Mac app checks, so either notice can appear even when the window looks unchanged. macOS 0.3.8 checks the window only; the 0.3.9 implementation also compares a focused Accessibility control when one is exposed at recording start. If no control is exposed then, or two fields reuse one identity, a same-window move can still paste into the new focus. Keep the original field and browser tab focused until insertion or a recovery notice appears. A failed input event does not prove that no text reached the field. If Presspeech says Couldn't paste — use Copy Last Transcript in 0.3.8, or Delivery uncertain in builds with the revised notice, inspect the destination before trying again. If text is missing, choose Copy Last Transcript when offered, but remove any partial text before pasting the full transcript. When Recent Transcripts is off, Presspeech has no in-app copy-recovery entry; correct or remove any partial text before dictating again. Use the target-app check to classify repeatable behavior safely.

The previous clipboard content is pasted

Update to macOS 0.3.8 or later first. Legacy 0.3.7 used an automatic restore timer that could race a slow target; turn Restore clipboard after paste off while you update. No fixed delay proves that another app consumed the paste.

In 0.3.8, automatic restoration and delay presets are retired. Keep Previous Clipboard for Manual Restore is a fresh opt-in and is off by default, even when the old automatic setting was enabled. It can retain an eligible complete previous clipboard in memory for up to five minutes, limited to 64 MB and 256 representations. Presspeech never offers a partial restore if either limit is exceeded or a representation is unavailable; the disabled restore row explains why nothing was kept. After verifying the latest dictated text arrived, choose Restore Previous Clipboard… and confirm. Cancel if the target is still waiting to read the clipboard. Expiry, copying something else, disabling the option or quitting discards the snapshot without an automatic rewrite.

If previous content still arrives on 0.3.8, remove it, leave the manual option off, recover with Copy Last Transcript if available, and retry only with non-sensitive text. Report the target app and attempt count with privacy-safe diagnostics; do not include either clipboard value or the dictated text.

System audio stays muted

Unmute the Mac in Control Center or Sound settings, then reopen Presspeech. The app uses a local watchdog and recovery marker to restore audio after an interrupted recording, and checks that marker again at launch. If it happens again, copy diagnostics before quitting.

Presspeech reports an unexpected exit

The notice means a local marker remained after the previous process ended; nothing was sent anywhere. Choose Copy Diagnostics or Open Log from the notice if you want to report what happened.

The notification icon is missing

Select the up-arrow at the right of the taskbar to check the notification-area overflow. You can also launch Presspeech again from the Start Menu. The existing process restores its open window, or opens Setup before first-run completion and Settings afterward.

The model is still preparing or failed

Before reopening, retrying, or selecting an uncached model in Windows 0.1.12: a new model request may send Hugging Face usage telemetry and an already-configured or saved token; custom routing can change its destination. A TLS-inspecting HTTPS proxy trusted by the client can read any token sent. If you want to avoid possible telemetry, are concerned about a token or custom route, or are unsure, do not start another download; wait until Windows 0.1.13 is published and installed. If a TLS-inspecting proxy is in use and its trust is unclear, do not launch while it is in use. Read the full Windows privacy decision before retrying.

If a download is already in progress and you have decided to continue, keep Setup open and wait for the selected model to report ready. A hotkey press while it is preparing is intentionally ignored. If Setup reports a failure, check the connection and available disk space first; on an NVIDIA system, also check the display driver. If you choose to make another model request after reading the warning above, use Retry Speech Model or select another model in Settings. An uncached alternative can also download files.

The microphone is silent or cannot open

Published Windows 0.1.12 checks the selected input during setup and when its selection changes; the retry action is Check Again. Upcoming 0.1.13 and builds containing this change leave the microphone closed until you choose Check Microphone in Setup or Settings. Settings checks the input shown in its picker before Save, but only Save changes the dictation input; it also links to Windows Microphone Privacy and Sound Input settings. If the check detects no input level, unmute the selected microphone, speak, and repeat with Check Again in 0.1.12 or Check Microphone in 0.1.13. If it cannot open, use the buttons for Windows Microphone Privacy and Sound Input settings. Turn on Microphone access, Let apps access your microphone, and Let desktop apps access your microphone, then confirm the input under Settings → System → Sound → Input. Standard Windows builds use the shared desktop-app control without a separate first-access prompt; some Insider Experimental builds add per-app desktop controls and prompts. If prompted, identify the requesting app before allowing it; an Unsigned or unfamiliar label alone is not proof of Presspeech. Review the shared controls before opening 0.1.12, or before choosing Check Microphone in 0.1.13. If a USB or Bluetooth input was disconnected, reconnect it and choose Check Again in 0.1.12 or Check Microphone in 0.1.13; Presspeech refreshes microphone discovery without an app restart and does not silently replace a specifically selected input. Before opening the microphone for each recording, it confirms that its cached Windows audio-device index still names the configured input, rejecting a stale entry when re-enumeration shows that indexes changed after reconnect or resume.

The hotkey does nothing

Confirm the model says ready and check the selected key in Setup or Settings. If Right Alt types @, €, or accented letters, your keyboard uses it as AltGr; Presspeech deliberately leaves that chord alone. Choose F8 or another available modifier or F-key and retry in Try Dictation.

If the key stops working after lock/unlock or sleep/resume but notification-area Dictate still works, choose Repair Global Hotkey in the menu, Setup, or Settings. Published 0.1.12 needs this manual action. Source builds preparing 0.1.13 attempt to replace the listener automatically after unlock or an interactive resume; they discard an active capture when the session locks or suspends. If reconnection fails or is deferred too long, use the same manual repair action after any dictation has finished.

Try Dictation works, but text is not inserted

Click the destination text field before pressing the hotkey and keep that window focused until transcription finishes. In published Windows 0.1.12, Transcript copied, not pasted means the finished transcript was copied at completion. Check the intended field first, then press Ctrl+V only if the clipboard still holds the complete transcript; a later copy can replace it without changing the notice. If unsure, check it in a blank, disposable text field before pasting into the intended field. If it has changed, do not paste the newer item as your dictation.

This fallback is used when focus changed, the original window could not be identified, or the target is running as administrator. Reopen an elevated target normally when possible; do not run Presspeech as administrator as a workaround. Upcoming 0.1.13 instead retains uncertain dictation for an explicit Delivery Recovery Copy or Discard; an already-known missing, unfocused, or elevated target does not replace the previous clipboard item. Use the target-app check to classify repeatable behavior safely.

Windows delivery recovery in upcoming 0.1.13

These controls are not included in Windows 0.1.12. Builds containing the new recovery actions keep a finished dictation in memory when clipboard access fails, a newer copy replaces it before paste, the original target cannot be used, or keyboard delivery becomes uncertain. If Windows cannot provide the target app's executable name, Presspeech also retains the dictation without replacing the prior clipboard item rather than guessing a local or remote paste route. New recording waits for review.

Check the intended field first: an input error can occur after a paste. The keyboard-accessible Delivery Recovery window opens without showing or copying the dictated words. Choose Copy for Manual Paste to copy explicitly, Discard Dictation to forget the recovery copy without changing the clipboard, or Leave Waiting to close the window while recording remains paused. Copy replaces the current clipboard item, including an image or formatted copy, and Presspeech does not restore it; choose Leave Waiting if you need to save that item first. A blocked hotkey or reopening Presspeech brings the window back; neither action copies automatically. Equivalent Review Undelivered Dictation…, Copy Undelivered Dictation (replaces clipboard), and Discard Undelivered Dictation commands remain in the notification-area menu.

Depending on where delivery became uncertain, some or all text may already be in the field or on the clipboard. If the clipboard changes during recovery, Presspeech keeps its recovery copy in process memory for another deliberate Copy or Discard. Exit forgets it. Sequence checks do not make the paste atomic or acknowledge that another application consumed the text.

Every 0.1.13 dictation clipboard item is marked for exclusion from Windows Clipboard History and Cloud Clipboard before transcript text is published. This does not remove the current item: deliberate Ctrl+V still works, and other local software or a third-party clipboard manager can still read it.

Playback stays muted

Unmute the affected output from Windows Quick Settings or Settings → System → Sound, then reopen Presspeech. Recordings normally restore each playback device to its previous mute state, including when the configured maximum length stops a recording. Copy diagnostics if the state is not restored.

Start with Windows will not enable

Choose Open Startup Settings, find Presspeech under Settings → Apps → Startup, and review the system setting. Then return to Setup or Settings and retry Finish Setup or Save. Finish Setup stays open if registration fails. In upcoming Windows 0.1.13, Set Up Later instead shows a warning and closes without completing setup; the guided setup returns on the next launch.