HyperYap is a privacy-first voice-to-text app that runs speech recognition locally on your machine. It uses NVIDIA Parakeet, collects no data, and gives you fast recording, automatic paste, configurable shortcuts, optional LLM cleanup, and terminal-friendly workflow helpers.
- What You Get
- Why This Model
- Install
- Shortcut Support
- Terminal-Friendly Workflows
- Platform Compatibility
- Requirements
- How It Works
- Configuration
- Build from Source
- Contributing
- Attribution
- License
One app for private, local transcription:
- Local speech-to-text powered by the English NVIDIA Parakeet TDT 0.6B v2 model through the smcleod INT8 ONNX conversion. No cloud, no internet after first install.
- Fast record-and-paste workflow that transcribes speech and inserts the result into the active app.
- Configurable shortcuts for recording, last transcript paste, command mode, cancellation, and LLM-assisted modes.
- Toggle-to-talk and push-to-talk recording modes.
- Optional LLM post-processing through local Ollama or a configured remote endpoint.
- Custom dictionary and formatting rules for names, commands, casing, and recurring phrases.
- Import/export settings for moving configurations between installs.
- Desktop builds for Windows, macOS Apple Silicon, macOS Intel, and Linux x86_64.
HyperYap is optimized for English-first local dictation. The default model is NVIDIA's Parakeet TDT 0.6B v2, packaged through the smcleod/parakeet-tdt-0.6b-v2-int8 ONNX INT8 conversion.
That choice is deliberate. HyperYap is built around fast, private, everyday English speech-to-text rather than broad multilingual coverage. The v2 Parakeet line is English-focused, the ONNX INT8 conversion fits the local desktop runtime, and the model can run fully offline after the first download. For users whose main need is English dictation into editors, terminals, chat apps, and coding tools, this keeps the app focused on the workflow it is meant to serve.
Multilingual Parakeet models are still valuable for multilingual transcription, but they are not the default direction for HyperYap right now. The project prioritizes English quality, local reliability, low-friction installation, and a fast record-and-paste loop.
Download the package for your platform from the latest release.
Run hyperyap_1.0.10_x64-setup.exe from the Releases page.
For the full workstation setup, use the PowerShell installer. It installs HyperYap, downloads the speech model, and configures the optional hotkey helper:
irm https://raw.githubusercontent.com/avalonreset/hyperyap/main/install.ps1 | iexOr clone and run locally:
git clone https://github.com/avalonreset/hyperyap.git
cd hyperyap
powershell -ExecutionPolicy Bypass -File install.ps1The PowerShell installer also removes old MURmure installations if present.
Download the matching DMG from the latest release:
HyperYap_aarch64_darwin.dmgfor Apple Silicon MacsHyperYap_x86_64_darwin.dmgfor Intel Macs
macOS requires Accessibility permission for global shortcuts.
Download HyperYap_amd64.AppImage or HyperYap_amd64.deb from the latest release.
On Debian/Ubuntu, you can also use:
curl -fsSL https://raw.githubusercontent.com/avalonreset/hyperyap/main/install.sh | shRun the installer again over an existing install. HyperYap upgrades in place without deleting your speech model or requiring a re-download. Settings are reset to the recommended defaults on each upgrade.
HyperYap records through configurable shortcuts. The exact binding can be changed from Settings.
| Shortcut | Action |
|---|---|
| Record shortcut | Start or stop recording |
Ctrl+Shift+Space |
Paste last transcript |
Ctrl+Alt+Space |
LLM-assisted recording |
Ctrl+Shift+X |
Command mode |
Ctrl+Shift+1 through Ctrl+Shift+4 |
LLM mode slots |
Escape |
Cancel recording |
The Windows workstation preset maps F13, CapsLock, and Mouse Back to the record shortcut, and maps Mouse Forward to Enter. Those mappings come from the optional hotkey helper and can be paused from its tray menu.
HyperYap can be used with any app that accepts pasted text. It is especially useful in editors, chat apps, terminals, issue trackers, and coding tools where fast dictation reduces context switching.
The optional hotkey helper adds terminal-aware behavior where supported:
- Smart copy, paste, and undo handling for terminal windows.
- Clipboard screenshot conversion into a saved PNG path before paste.
- Bounded paste undo for recent HyperYap-managed terminal inserts.
Supported terminal process names for the helper:
| Terminal | Process |
|---|---|
| WezTerm | wezterm-gui.exe |
| Windows Terminal | windowsterminal.exe |
| PowerShell | powershell.exe |
| PowerShell 7+ | pwsh.exe |
| Command Prompt | cmd.exe |
| Alacritty | alacritty.exe |
| ConEmu | conemu.exe / conemu64.exe |
| Hyper | hyper.exe |
| Git Bash (mintty) | mintty.exe |
| Tabby | tabby.exe |
| Warp | warp.exe |
| MobaXterm | mobaxterm.exe |
The core app is built for Windows, macOS, and Linux. Global shortcut behavior is implemented separately per platform, so support depends on the operating system permissions and desktop environment.
| Capability | Windows | macOS | Linux |
|---|---|---|---|
| Local Parakeet transcription | Supported | Supported | Supported |
| Configurable recording shortcuts | Supported | Supported with Accessibility permission | Supported where the desktop/session allows global input hooks |
| Automatic paste into active app | Supported | Supported | Supported |
| CapsLock and mouse side-button remapping | Supported through the optional hotkey helper | Not currently bundled | Not currently bundled |
| Smart terminal copy/paste/undo helper | Supported through the optional hotkey helper | Not currently bundled | Not currently bundled |
| PowerShell one-line workstation installer | Supported | Not applicable | Not applicable |
In short: the app should work as a local voice-to-text tool on macOS and Linux, but the CapsLock remapper and smart terminal helper are currently Windows-only. macOS users must grant Accessibility permission. Linux global shortcuts may vary by distribution, window manager, and Wayland/X11 session.
- Windows 10+, macOS, or Linux x86_64
- A microphone
- ~700MB disk space for the speech model
- Internet connection for first launch model download
- Press your configured record shortcut.
- Speak naturally into your microphone.
- Press the shortcut again to stop recording.
- HyperYap transcribes locally using the Parakeet TDT model.
- The transcription is automatically pasted into the active window.
All processing happens on your machine. Audio never leaves your computer. The speech model runs entirely offline after the initial download.
HyperYap works out of the box with minimal setup. All settings can be changed from the app's Settings page.
| Setting | Default | Description |
|---|---|---|
| Record mode | Toggle-to-talk | Press once to start, press again to stop. Can be changed to push-to-talk. |
| Record shortcut | Configurable | Remap to any supported key or key combination. |
| Language | English | Optimized for English local dictation. |
| Overlay | Bottom of screen | Recording indicator position. Can be set to top, bottom, or hidden. |
| LLM Connect | Disabled | Post-process transcriptions with a local LLM through Ollama or a configured remote endpoint. |
| HTTP API | Disabled | Local API on localhost for external tool integration. |
| Sound feedback | Enabled | Audio cues when recording starts and stops. |
| Copy to clipboard | Disabled | Optionally keep transcriptions in the clipboard. |
Settings are stored in %APPDATA%/com.avalonreset.hyperyap/settings.json on Windows. Each install or upgrade resets settings to the recommended defaults. If you need to preserve custom settings across upgrades, back up this file before updating.
All in-app shortcuts can be remapped from the Settings page. Platform-level remapping such as CapsLock or mouse side buttons is handled by the optional hotkey helper where available.
pnpm install
# Build the optional hotkey helper before packaging the Windows installer
cd hotkeys && cargo build --release && cd ..
pnpm tauri dev # development
pnpm tauri build # production buildRequires: Node.js 18+, Rust, pnpm, Tauri prerequisites
Download the default Parakeet ONNX model into resources/parakeet-tdt-0.6b-v2-smcleod-int8/:
powershell -ExecutionPolicy Bypass -File .github/scripts/download-parakeet-model.ps1 -Destination resourcesOn macOS/Linux:
bash .github/scripts/download-parakeet-model.sh resourcesContributions are welcome. See CONTRIBUTING.md for development setup, PR workflow, and coding guidelines.
Please read the Code of Conduct before contributing.
HyperYap's voice engine is a modified version of MURmure by Kieirra. Full credit to the original author for building an excellent local speech-to-text application.
Powered by NVIDIA's Parakeet TDT 0.6B v2 English speech recognition model, using the smcleod/parakeet-tdt-0.6b-v2-int8 ONNX INT8 conversion.
The voice engine is licensed under AGPL-3.0. See NOTICE for full attribution details.
