Speed of Sound

Voice typing for the Linux desktop

io.speedofsound.SpeedOfSound

Speed of Sound is a voice-typing app for the Linux desktop.

By default, transcription runs entirely on-device, so no audio ever leaves your machine. The app ships with a built-in multilingual Whisper model and supports additional models from the Whisper, Parakeet, and Canary families to improve accuracy.

Offline, private transcription with popular open speech models (Whisper, Parakeet, Canary)
Multiple activation options: click the in-app button or use a global keyboard shortcut
Types directly into any app via XDG Desktop Portals (supports X11 and Wayland)
Optional text polishing with LLMs, supports self-hosted services like Ollama, vLLM, and llama.cpp
Optional cloud providers supported for resource-constrained hardware environments

Please visit speedofsound.io/getting-started after installation.

Website ↗Help ↗Donate ↗Source code ↗Report an issue ↗Contribute ↗Subscribe via RSS

Screenshots

Voice typing into the default text editor

Add personal context and custom words to improve accuracy

Observed history

App	Change	Version	When ↓

No observed changes yet.

The initial import establishes a baseline. Future version changes will appear here.

Browse the app catalogue →

Upstream releases

Release notes supplied by the developer. These dates are not Flathub package publication times.

0.14.09 May 2026

Added clipboard output as an alternative to keyboard simulation — useful when the XDG Remote Desktop Portal is unavailable or restricted. Tap the language label in the main window to quickly toggle between your primary and secondary language without opening Preferences. Updated the Gradle wrapper to 9.4.1 to match the Flatpak SDK extension.

0.13.021 Apr 2026

Added a system tray icon with Start/Stop Listening, Open App, and Quit actions for quick control without opening the main window. New Disable Thinking option for LLM text providers reduces latency and cost on models that support reasoning, and a Test Connection button lets you verify credentials before saving. Fixed the app auto-quitting after a few seconds when "Stay hidden on activation" was enabled. Fixed a native resource leak where the Sherpa ONNX stream was not released on transcription failure. Updated Java-GI to 0.15.0 and Stargate to 0.5.0.

0.12.011 Apr 2026

Added support for NVIDIA Canary and Parakeet ASR model families alongside Whisper, giving more options for on-device speech recognition across multiple languages. New "Stay hidden on activation" setting lets you add the app to system startup without showing the main window. The app now detects when the XDG Remote Desktop Portal is not supported and shows a clear banner instead of failing silently. Fixed typing error propagation so non-Latin script failures surface as notifications instead of being swallowed. Fixed a bug where changing the secondary language in Preferences didn't update the main window. Updated dependencies including Sherpa ONNX, Kotlin, and cloud SDKs.

0.11.029 Mar 2026

Added Start and Stop buttons to the main window for easier control without a keyboard shortcut. Fixed Unicode and CJK character support so text in Arabic, Chinese, Japanese, Korean, and other non-Latin scripts now types correctly. The app now exits immediately with a clear error on fatal startup failures instead of hanging silently. Icons are now bundled as compiled GLib resources for reliable display across all distribution formats.