Download TRWhisper
TRWhisper is a Windows 11 dictation program that converts speech into text using a local speech recognition engine. Pressing and holding the Right Ctrl key captures microphone input, and releasing the key produces transcribed text that is copied to the clipboard and pasted into the active window. The tool runs without sending audio to remote servers during normal operation. It supports optional graphics card acceleration for faster results and includes a tray menu for switching models and reviewing recent transcripts.
The software uses the whisper.cpp engine and the Silero voice activity detection model to process speech locally. An optional cleanup mode can forward text to a language model API for removal of filler words and punctuation correction. A floating status indicator shows recording and processing states, and all transcripts are saved to monthly log files in the user profile folder. The installer bundles the application, a CPU engine, and a silence detection model, with optional downloads for GPU support and larger recognition models.
TRWhisper targets users who need speech-to-text functionality without cloud processing. It is written in C# for the .NET 9 platform and distributed as a single-file setup executable. The application registers a global keyboard hook for its push-to-talk hotkey and uses the Win32 SendInput API for automatic pasting. A hallucination filter removes known non-speech patterns from output. The project is maintained on GitHub with documentation in English and Turkish, reflecting its original design focus on Turkish dictation.
Use Cases for TRWhisper
| Low-level keyboard hook captures Right Ctrl globally | Hold-to-record release-to-transcribe without window switching |
| Clipboard placement followed by simulated Ctrl+V insertion | Text remains available for subsequent manual pasting |
| Bundled whisper.cpp engine performs recognition on local machine | Temporary WAV files deleted immediately after processing |
| CUDA build offloads inference to compatible NVIDIA hardware | Automatic CPU fallback when no usable GPU is detected |
| Tray submenu switches Whisper models without restart | Absent models greyed out, unavailable options flagged |
| Silero VAD excludes non-speech segments before transcription | Silent hotkey presses produce no phantom output text |
| Pattern matching removes known hallucination artifacts from transcripts | Line-level filter preserves legitimate dictation content |
| On-screen pill overlay reflects recording and processing states | Cancel and early-finish controls with result copy button |
| Optional language model cleanup via Gemini or OpenAI endpoints | Filler removal and punctuation correction requiring API credentials |
| Tray icon colour encodes idle, recording and transcribing states | Context menu exposes models, history, folders and settings |
| Monthly Markdown log of completed transcriptions with timestamps | Separate diagnostic log with temporary audio cleanup |
| Installer bundles VAD model and CPU engine with optional GPU package | SHA-256 verified downloads without administrator privileges |