# Dictata

> 100% local, system-wide voice dictation for Windows built in Rust, using whisper.cpp and optional ONNX backends to transcribe speech and paste text into any active application without sending data off-device.

Dictata is a Windows desktop application written in native Rust that lets users dictate text into any application using a global hotkey — no cloud, no network calls, no data leaving the machine. Created by Antoine Chatry and released as v0.1.0, it runs transcription locally through the companion [Dictata Engine](https://github.com/AntoineChatry/DictataEngine), which wraps whisper.cpp by default and supports several optional ONNX backends.

## What It Is

Dictata is a local-first speech-to-text dictation tool for Windows 10/11. It sits in the system tray, listens for a configurable global hotkey (default `Ctrl+Alt+Space`), records audio, transcribes it entirely on-device, and pastes the result into whatever application has focus. The core transcription engine is whisper.cpp (via whisper-rs), with optional ONNX backends — NVIDIA Parakeet, SenseVoice, Moonshine, and Zipformer — each enabled at build time via Cargo features.

## Architecture and Backends

Dictata is structured as a multi-module Rust application. Key modules include audio capture (`audio.rs` using cpal with WASAPI loopback), a transcription facade (`transcriber.rs`), a multi-backend ASR engine layer (`engine/`), streaming/continuous mode (`streaming.rs`), and an egui-based settings UI with eight pages. The backend is selected automatically from the shape of the model file: a ggml `.bin` file routes to whisper, while an ONNX bundle directory routes to its matching backend. GPU acceleration is supported via Vulkan (AMD, Intel, NVIDIA) or CUDA, with a CPU-only build path available.

## Key Features

- **Toggle or push-to-talk hotkey** with automatic paste into the active window
- **Continuous/streaming mode**: text is inserted progressively at every detected pause
- **Voice Activity Detection (VAD)**: optional silence skipping to reduce compute and hallucinations
- **Audio sources**: microphone, system audio (WASAPI loopback), or a mix of both for meeting transcription
- **Output post-processing**: raw text or cleanup/formatting through a local OpenAI-compatible LLM (email, message, list, and custom prompts)
- **File transcription**: any audio/video format handled by ffmpeg, accessible from the History page
- **Built-in model library**: download, delete, and hardware-aware recommendations for ggml and ONNX models; HuggingFace search by URL, repo, or keyword
- **Custom vocabulary and replacements** injected as the initial prompt
- **Multilingual UI**: French, English, and Spanish; dark theme; system tray icon; transcription history

## Platform and Requirements

Dictata targets Windows 10/11 exclusively in its current release. Linux support is explicitly noted as incomplete — a `LINUX.md` file documents what works and what does not. The only runtime dependency for file transcription is `ffmpeg` in `PATH`. GPU builds require the Vulkan SDK and Visual Studio Build Tools; CPU-only builds have no additional dependencies. The compiled binary is self-contained with no DLLs to ship alongside it.

## Update: v0.1.0

The repository was created in June 2026 and last pushed in August 2026. The current release is **v0.1.0**, with a `CHANGELOG.md` tracking changes. The project has 7 GitHub stars and 0 forks at the time of indexing, indicating it is an early-stage personal/indie release. The README describes 83 unit tests covering config, resampling, mixing, modes, settings logic, hardware rating, HuggingFace query parsing, history retention, and backend detection.

## Licensing Model

Dictata is released under the **MIT License with the Commons Clause** condition. This means the source code is freely available to use, copy, modify, distribute, and self-host for personal or any other purpose, but commercial resale — including hosting, SaaS, or paid support services whose value derives substantially from the software — is prohibited. Because the Commons Clause restricts commercial use, the license is not OSI-approved open source despite the MIT base.

## Features
- Global hotkey dictation (toggle or push-to-talk)
- Automatic paste into active application
- Continuous/streaming mode with progressive text insertion
- Voice Activity Detection (VAD) for silence skipping
- Vulkan GPU transcription (AMD/Intel/NVIDIA) or CPU
- Multi-backend ASR engine: whisper.cpp, Parakeet, SenseVoice, Moonshine, Zipformer
- Audio sources: microphone, system audio (WASAPI loopback), or mix
- Output post-processing via local OpenAI-compatible LLM
- File transcription (audio/video via ffmpeg)
- Built-in ggml and ONNX model library with hardware-aware recommendations
- HuggingFace model search by URL, repo, or keyword
- Custom vocabulary and replacements as initial prompt
- Floating dock with waveform (configurable size, opacity, position)
- Multilingual UI: French, English, Spanish
- Dark theme, system tray icon, transcription history

## Integrations
whisper.cpp, ONNX Runtime, NVIDIA Parakeet, SenseVoice, Moonshine, Zipformer, ffmpeg, HuggingFace, Vulkan SDK, WASAPI, cpal, egui, Local OpenAI-compatible LLM

## Platforms
WINDOWS, LINUX, API, CLI

## Pricing
Open Source

## Version
v0.1.0

## Links
- Website: https://github.com/AntoineChatry/Dictata
- Documentation: https://github.com/AntoineChatry/Dictata/blob/master/README.md
- Repository: https://github.com/AntoineChatry/Dictata
- EveryDev.ai: https://www.everydev.ai/tools/dictata