# VoiceTray

VoiceTray is a free, open-source Windows tray dictation app. Hold F9, speak, and release to run local faster-whisper transcription, guarded local cleanup, local SQLite history, and safe insertion into the focused application.

Website: https://voicetray-offline.vercel.app/
Simulation: https://voicetray-offline.vercel.app/demo/
Repository: https://github.com/tanzir71/voice_tray

The public simulation is canned browser state. It requests no microphone access and performs no audio capture, STT, network request, or foreground insertion.

## From source

pip install -r requirements.txt
python -B -m pytest tests\ -q
python -B -m voicetray.eval
pythonw -m voicetray

Windows and a microphone are required for the real app. tools/build.ps1 creates the one-folder packaged application.

## Dictation path

1. F9 hold or lock mode records locally with VAD and a floating pill.
2. faster-whisper transcribes using base, small, or medium.
3. Deterministic rules remove fillers/repetitions, apply spoken punctuation/lists, and handle backtracking.
4. An optional local GGUF LLM can polish text; risky edits fail validation and fall back to rules.
5. Per-app profiles, local glossary, and snippets refine output.
6. Raw and clean text are saved in local SQLite before safe foreground insertion.

The runtime dictation path has zero network calls, telemetry, or cloud fallback. Network is used only for explicit model downloads. VoiceTray is Windows-only in v1, has no command mode, and local accuracy/speed depends on model size and hardware.