Changelog
Every SpeakText release, improvement, and fix from CHANGELOG.md.
Truly live typing, and a switch to turn automatic punctuation off.
Added
- Automatic punctuation can now be turned off with the new Punctuation switch under your shortcut. With it off, SpeakText types your words only: no periods, commas, or capital letters added for you. "I" and names keep their capitals. The switch applies instantly, with no reload of the speech engine.
Changed
- Words now appear as you speak, even during long sentences. Previously a misheard first word could hold back all the text until you stopped talking.
- Fixed occasional repeated phrases in live text.
- Thinking pauses still never end your sentence, and short greetings like "Hey guys." keep their period.
- Newly connected microphones (like a Bluetooth headset) now appear as soon as you open the Mic menu, without restarting SpeakText.
- SpeakText remembers your chosen microphone by name, so plugging in another device can't switch you to the wrong mic. If your mic is disconnected, SpeakText uses the Windows default and switches back automatically when it returns.
- Cleaner window: the minimize button is gone (closing hides SpeakText to the tray), and the close button no longer looks like it quits the app.
Natural punctuation: pause to think without breaking your sentences.
Changed
- Pausing to think no longer ends your sentence. SpeakText waits to see what you say next before adding a period, so you get full, properly punctuated sentences instead of short fragments.
- SpeakText waits longer before finishing a phrase when you stop mid-thought (for example after "and" or "the"), so you aren't cut off.
- Each new phrase is transcribed with the context of what you've already said, which keeps capitalization and punctuation consistent across pauses.
- New "Auto" speech model (now the default) picks the most accurate model your computer can run in real time.
- Upgraded the speech engine to RealtimeSTT 1.1.2, with faster voice detection.
Faster starts, clearer downloads, and more dependable dictation.
Changed
- Speech setup now reuses a recently verified installation, so repeat launches reach dictation faster.
- Model downloads now show clear preparing, downloading, ready, interrupted, and retry states.
- Live speech previews stay visible in SpeakText while only stable words are typed into the active app.
- Push-to-talk responds immediately when held, without waiting for voice detection.
- Time-saved and dictation totals are preserved more reliably across updates and repairs.
- The website changelog is included in the page itself, so the full history appears immediately instead of getting stuck on “Loading…”.
More reliable push-to-talk and smoother long-form dictation.
Changed
- Added a dedicated Windows-wide push-to-talk system that works consistently across normal desktop apps.
- Improved how live words are revised, reducing repeated phrases and awkward corrections during longer thoughts.
- Improved microphone discovery and clearly falls back to the Windows default if a saved device is unavailable.
- Tuned speech timing so consecutive phrases feel quicker and more natural.
Safer updates and a more polished everyday tray experience.
Changed
- Updates are downloaded over secure connections and verified before installation.
- The tray menu now gives clearer access to settings, pause/resume, the active shortcut, the website, and quit.
- Shortcut capture and normalization now support a wider range of Windows keys more reliably.
- Microphone levels, partial transcripts, and settings updates reach the interface more smoothly.
- The website now sets accurate expectations that SpeakText currently focuses on English voice typing for Windows.
Stronger foundations for a product customers can install and trust.
Changed
- Restored and hardened the portable and single-file Windows installers.
- Reduced background processing used by the microphone level display.
- Added automated release checks to catch app, website, and speech-engine regressions earlier.
- Strengthened website security and caching for safer, faster browsing.
- Published SpeakText under the MIT license for greater transparency.
A clearer look at how SpeakText fits into a real workflow before purchasing.
Added
- Added an interactive preview showing SpeakText typing directly into email, browser, developer, and document workflows.
- Added a full settings preview covering speech models, shortcuts, microphones, typing behavior, and time-saved statistics.
- Added a dedicated Spokenly comparison alongside expanded alternative-app guidance.
Changed
- Reworked product pages to explain local cursor dictation, Windows focus, and lifetime pricing more clearly.
Cleaner final text and a more useful purchase experience.
Changed
- Improved the handoff from live words to the final transcript so corrections are less likely to duplicate text.
- Made back-to-back recordings more dependable when speech starts again quickly.
- Added localized US pricing detection and direct access to the customer portal from the website.
- Improved the prerendered website so product pages are complete and discoverable before JavaScript loads.
Microphone choice and better control over the dictation experience.
Added
- Added a microphone picker so users can choose a specific input or keep the Windows default.
- Added clearer setup and recording feedback in the settings window.
Changed
- Improved recovery between recordings so SpeakText is ready for the next phrase without a restart.
- Extended continuous dictation sessions while keeping live text responsive.
- Improved the stability of word corrections near the beginning of an utterance.
Quicker pause-and-resume behavior for everyday use.
Changed
- Keeps the speech model warm while paused, allowing dictation to resume faster.
- Improved continuous recording transitions so completed speech can be processed while listening continues.
- Added clearer listening and recording states plus an easy way to copy setup details when support is needed.
Substantially steadier realtime typing.
Changed
- Improved live word stabilization to reduce duplicated, missing, or unexpectedly replaced text.
- Stops an utterance safely if focus moves to another typing target, helping prevent words from landing in the wrong app.
- Improved long-dictation handling by processing speech in manageable segments without interrupting the flow.
- Expanded recognition of downloaded speech models and made pausing less resource-intensive.
More transparent first-run setup and easier speech-model selection.
Added
- Added visible setup progress for backend checks, speech-engine installation, model downloads, and model loading.
- Added a setup log panel for users who want more detail or need help troubleshooting.
- Added clearer model descriptions and automatically hides GPU-only choices on CPU systems.
Changed
- Improved background model downloads and setup recovery so the app stays informative throughout first launch.
All commits since [36f1a70]...
Added
- Model download script (download_model.py) for manual model management
- Screen API integration for enhanced window management
- Extended speech model support (tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large-v3, turbo, distil-small.en, distil-medium.en, distil-large-v2, distil-large-v3)
- Productivity tracking data (totalWords, totalDictationSeconds) with shared runtime storage
- Worker health monitoring (workerStartedAt, workerLastSignalAt, workerLastAudioLevelAt)
- "transcribing" state to WorkerState for better status tracking
- Worker prewarming capability
- SSG/SSR prerender script for website
- Copy-static script for website assets
- Sync-changelog script for website
- Privacy page to website navigation
- New entry-server.tsx for SSR
Changed
- Major refactoring of bun/index.ts with improved worker management and productivity tracking
- Enhanced window hiding from Windows taskbar
- Updated push-update.ts script with extensive changes for multi-channel releases
- Website UI styling with new light-button class and comprehensive redesign
- Removed startup paused behavior for login launches (always starts recording)
- Improved version metadata resolution (metadata.json and version.json support)
- Refined STT worker text normalization and stability
- Updated Python STT worker with improved streaming and finalization logic
- Enhanced model caching and download handling
- Updated nginx.conf for better routing
- Website Dockerfile improvements
Fixed
- Various STT worker state management improvements
- Multiple STT worker initialization refinements
- Worker lifecycle tracking
- Improved text streaming stability and word boundary handling
- Settings persistence issues
Removed
- Old manifest URL format (replaced with update.json)
[36f1a70 through a3fe6cf]
Release build for the latest SpeakText app download.
Changed
- Built new canary and stable app packages.
- Updated the Polar customer download package.
Release build for the latest SpeakText app download.
Changed
- Built new canary and stable app packages.
- Updated the Polar customer download package.
Release build for the latest SpeakText app download.
Changed
- Built new canary and stable app packages.
- Updated the Polar customer download package.
Release build for the latest SpeakText app download.
Changed
- Built new canary and stable app packages.
- Updated the Polar customer download package.
All commits since [36f1a70]...
Added
- Productivity tracking system (totalWords, totalDictationSeconds) with shared runtime storage
- GitHub Releases API integration for update checks (replacing custom manifest URL)
- Worker health monitoring with timestamps (workerStartedAt, workerLastSignalAt, workerLastAudioLevelAt)
- "transcribing" state to WorkerState enum
- Worker prewarming capability
- New "Privacy" page to website navigation
- Enhanced window hiding from Windows taskbar
Changed
- Major refactoring of bun/index.ts with improved worker management
- Updated push-update.ts script with extensive changes
- Website UI styling with new light-button class
- Removed startup paused behavior for login launches
- nginx.conf cleanup
Fixed
- Various STT worker state management improvements
- Multiple STT worker initialization refinements
- Worker lifecycle tracking
[36f1a70 through b977b38]
All commits since [e463058]...
Added
- C# .NET 8 bootstrapper (SpeakTextBootstrapper) for Python/GPU dependency installation
- Single-file executable packaging (build-single-exe.mjs, package-portable-win.mjs)
- Auto-update system with versioned releases (v0.1.0 through v0.1.6)
- Windows post-build icon script (postbuild-win-icons.mjs)
- Speakflow branding assets (PNG logo)
- TypeScript type definitions for Three.js integration
- WebSocket communication layer for STT worker integration
Changed
- Enhanced tray app UI with improved styling and visual branding
- Updated STT worker with better state management and error handling
- Improved bootstrap-gpu.ps1 with CUDA verification
- Refined website App.tsx with content and styling updates
- Updated push-update.ts for better versioning and artifact handling
- Enhanced docker-compose and Dockerfile configurations
Fixed
- Various UI state synchronization issues
- STT worker initialization race conditions
- Bootstrap script path handling
[e463058 through 36f1a70]
Added
- Electrobun tray application with real-time speech-to-text
- Python STT worker with RealtimeSTT integration
- GPU bootstrap script for CUDA-enabled PyTorch setup
- SolidJS + Tailwind v4 marketing website
- Docker deployment support (Dockerfile, docker-compose.yml, nginx config)
- TypeScript type definitions
- Deployment scripts (deploy-release.ps1, push-update.ps1)
- Environment configuration (.env)
- Static updates directory for website
- Logo assets (Speakflow.png, logo.png)
- Python CUDA verification script
Changed
- Updated README with project layout and quick start guide
- Enhanced tray app UI with styling (index.css, index.tsx)
- Improved STT worker with better error handling and state management
- Refined website App.tsx with content and styling
- Updated gitignore files for both speaktext-tray and website
Fixed
- Pycache files tracked correctly
[Initial]: e463058
One price. Lifetime access.
No subscriptions. No tiers. Just a single purchase that never expires.
Single Lifetime License
€19.99 EUR
One-time payment. Free updates forever.
Runs locally on your Windows machine. GPU is recommended for fast realtime dictation; CPU also works, with speed depending on your hardware and model choice.
- ―Realtime dictation at cursor
- —English transcription in the current release
- ―Local-first — no data leaves your machine
- ―Works across all your applications
- ―GPU recommended for realtime local transcription; CPU also supported
- ―Lifetime updates and support
Secure payment. No subscriptions. No hidden fees.