Repository navigation
Releases
Releases · lman80/SwipeSpeak-app
SwipeSpeak 3.3.1
SwipeSpeak 3.3.0
Added
- TalkTalk as a wireless talk key. In SwipeSpeak mode, TalkTalk (the
keychain recorder) works like your talk key: press its button and
SwipeSpeak starts dictating at once, with TalkTalk's microphone doing the
listening; let go and your words are typed, with the same modes, cleanup
and History as the keyboard. Pair it once in Dictation › TalkTalk (listed
with General › Show advanced settings until a TalkTalk is paired; type the
6 digits shown on TalkTalk when your Mac asks); after that it connects by
itself whenever it's nearby and in SwipeSpeak mode. The page also turns SwipeSpeak mode on or
off, sets when it turns itself off, and switches Tap to talk. TalkTalk
keeps every dictation until SwipeSpeak has it safe, so nothing is lost if
Bluetooth drops (#32).
Changed
- A new look (Liquid Glass): floating glass sidebar, white cards,
bigger icons and a simpler Home. - A short menu bar menu (#38): status, Start/Stop Dictation, Paste
Again, Read Aloud on/off, Open SwipeSpeak, Send Feedback and Quit.
Everything else is in the main window. - A feedback button floats in the corner of the main window (#40).
- Setup asks about Live Preview (#33): a page in the Dictation track
turns it on and asks for Speech Recognition. - Long recordings always transcribe (#31): the "Transcribe N minutes?"
prompt is gone. The "Still recording?" reminder stays.
Fixed
- TalkTalk could freeze dictation (#39): a TalkTalk recording now ends
by itself when audio stops for 3 s or a new press arrives, Escape cancels
it, and the keyboard talk key always works (it stops TalkTalk's
recording). Refused Bluetooth writes are retried. - Lag on the TalkTalk page (#41): the 3D model draws only while it's
visible and moving, and no longer floods the log with render errors. - Nothing can lock the talk key anymore: every dictation step has a
time limit; if one hangs, it's cancelled and its audio stays in History. - Garbled letters instead of your words (#28): browsers, web apps and
long text are always pasted; typing sends one whole character per key
with no modifiers and waits for held keys to be released. - Made-up words (#37): "thank you", "yeah" and random words over
silence are removed using Whisper's own confidence; silence at the ends
is trimmed before upload. - Long dictations (#29): pieces sent while you talk are cut only at
pauses, each gets the text before it as context, and repeated words at
the joins are removed. - Offline tags (#30): "[BLANK_AUDIO]", "[laughing]", "(music)" and
similar are removed. - Apple speech model "Download" (#34) is offered only when it isn't on
the Mac already.
SwipeSpeak 3.2.2
Fixed
- The talk key no longer types itself or stops mid-dictation. If
SwipeSpeak was busy for a moment, a press of the talk key could go through
to your app as a typed\, or a release could be missed. The talk keys now
run on their own and never wait for the rest of the app. - When the talk key can't be heard, SwipeSpeak says why. A password box
in focus, Terminal's Secure Keyboard Entry, or macOS holding on after an
unlock hides keys like\from every app. Home and the menu bar now name
the app and say how to get the key back. - Deleting the selection while you talk makes it a normal dictation.
Select text, start talking (Edit Mode), then delete it: your words are
typed at the cursor instead. - "Learn this word?" shows up right after you fix a word, not at your
next dictation. - Downloading an update no longer slows the app down, and Check Now
shows its result in General without an extra alert.
New
- More speech engines in Dictation › Transcription (with your own key):
Grok Voice Transcribe 2.0 (xAI) and GPT-4o Transcribe / mini (OpenAI),
next to Groq's Whispers. - Feedback can now take a screenshot of another app's window, and has
Feedback / Idea types.
SwipeSpeak 3.2.1
New
- SwipeSpeak Pro: no key to set up. The setup now asks how SwipeSpeak
should hear you: SwipeSpeak Pro (free for now, works right away) or
Your own key (a free Groq key you paste once). You can switch any
time in Dictation › Transcription. ✨ summaries use the same choice.
If you already use your own key, nothing changes.
Fixed
- Accessibility that stays "Waiting for the switch…" (from 3.2.0): Open
Settings clears your Mac's old entry for SwipeSpeak and adds this copy
back, so the switch you turn on works.
SwipeSpeak 3.2.0
New
- No key to type. SwipeSpeak now comes with a speech key built in, so
Dictation and ✨ summaries work right after setup. The Setup Guide skips
the "Add your speech key" page. You can still paste your own Groq key, and
it always takes priority.
Fixed
- Accessibility that stays "Waiting for the switch…". When your Mac
remembered an older copy of SwipeSpeak, its switch looked on but did
nothing. Open Settings now clears that old entry and adds this copy back,
so the switch you turn on works. "Switch is on but still waiting?" also
has Fix It and Restart buttons.
SwipeSpeak 3.1.9
Fixed
- "Read the latest AI reply" no longer flashes SwipeSpeak's window.
To have macOS read the reply, SwipeSpeak briefly takes focus with an
invisible window, and that also pulled its main window in front of your
app for about a second. The main window now stays where it was. - Circle to skip says why it can't skip (#19). Skipping presses the
buttons on macOS's reading controller; when Spoken Content is set to never
show it, the pointer now shows "set Show controller to Automatically"
instead of a bare "!". Skip also finds the controller through
Accessibility first, not only through the window list.
SwipeSpeak 3.1.8
Fixed
- "Read the latest AI reply" reads the whole reply (#18). It found the
reply but stopped after about a second on Macs with Spoken Content's
"Highlight content" on: SwipeSpeak closed the hidden window holding the
text as soon as reading began. The window now stays (invisible and
click-through) until the read ends. Listen-only ✨ summaries had the same
problem. - Circle to skip finds the reading controller's buttons more reliably
(#19), and a failed skip now notes why in the feedback log.
SwipeSpeak 3.1.7
Fixed
- Long dictations no longer crash the app (#16, #17). Since 3.1.6, a
cloud dictation longer than about a minute crashed SwipeSpeak: every bit of
recorded audio made the hand-off to "Send audio while I talk" one step
deeper until it ran out of room. - 🔊 no longer reads a split second and stops (#15). When Spoken
Content's controller is hidden, SwipeSpeak guessed how long a read takes;
at faster speaking rates the guess ran long, so a new read was started and
at once stopped, again on every try. It now checks whether Spoken Content
is actually playing. - "Read the latest AI reply" finds replies in T3 Code (#14), and on
chatgpt.com by its "ChatGPT said" labels. - Bigger sidebar icons that always fit inside their squares (#13).
SwipeSpeak 3.1.6
Added
- Updates install themselves. A new version downloads in the background
and installs when the Mac has been idle for two minutes and nothing is
recording or reading (or when you quit). General › Updates › "Install
updates automatically" turns it off. - Read the latest AI reply. Read Aloud › Reading › Shortcut: one key reads
the newest reply in T3 Code, ChatGPT, Claude or Gemini (app or browser) in
the Siri voice. Press again, or shake, to stop. - Faster long dictations. With the cloud, audio is sent at natural pauses
while you talk, so a long dictation is ready almost as fast as a short one
(Transcription › Advanced › "Send audio while I talk"). - Large v3 Turbo speech model option for Groq (faster; Large v3 stays the
default). - "Didn't paste?" pill. When a dictation clearly had nowhere to go, a pill
says which key pastes it again (#11).
Fixed
- Shake to stop works with gentler and trackpad shakes (#12).
- Send Feedback opens in front of the app you're using, on your current
desktop (#12). - Update checks run even before Accessibility is allowed.
SwipeSpeak 3.1.4
- Behind the scenes: gets SwipeSpeak ready for the next update, which Apple has checked (notarized), so downloads from the website open like any other app.