DictateAnywhere

High-performance, privacy-first hybrid voice dictation for Windows and macOS.
Injects transcribed text directly at your cursor in any active application using offline local faster-whisper (GPU/CPU) or active Cloud STT engines (Google Gemini, Microsoft Azure, Sarvam AI).

📥 Download Releases • 📖 Features • 🚀 Quick Start • 🛠️ Local Development

---

Overview

DictateAnywhere is an ultra-fast, privacy-first voice dictation application built with Tauri v2, Rust, React, and a Python Sidecar. It runs quietly in your system tray and types whatever you speak directly into any active input field across your operating system.

Designed for doctors, writers, coders, and power users, DictateAnywhere combines local GPU acceleration (NVIDIA RTX Tensor Cores) with leading cloud models to provide zero-lag, studio-grade speech-to-text.

---

Features

Feature	Details
Zero-Latency Capture	Instantaneous recording start with cached audio device discovery — zero delay upon pressing your hotkey.
Smart Silence Stripping	Integrated VAD pre-trimming with safety padding eliminates dead air and prevents Whisper repetition or hallucination loops.
NVIDIA GPU Acceleration	Full CUDA Tensor Core optimization (automatic float16 selection for RTX 5060, 40-series, and 30-series GPUs) with graceful CPU fallback.
Hybrid STT Architecture	Offline faster-whisper (GPU/CPU) with automatic cloud fallback on error.
Active Cloud STT Selector	Switch between Google Gemini Flash, Microsoft Azure Speech, or Sarvam AI with one click.
Disfluency & Filler Removal	Automatically purges hesitation sounds ("um", "uh", "hmm", "ah", "er") across all engines.
AI Text Polishing	Real-time LLM refinement via local Ollama or cloud models (Google Gemini, OpenRouter, Groq). Fix grammar, format, or apply custom prompts on the fly.
HTTP Connection Pooling	Persistent keep-alive sessions eliminate TCP/TLS handshake latency on cloud STT and LLM polish requests.
Floating Mic Button	Draggable, always-on-top toggle widget with recording state indicators and countdown sweep ring.
Global Hotkeys	Fully customizable keyboard shortcuts for dictation (default Ctrl+Alt+D / Cmd+Alt+D) and LLM polishing (Ctrl+Alt+P).
Hardened Security	Native Win32 asynchronous file dialogs via rfd, strict Tauri Content Security Policy (CSP), and header-authenticated API keys (x-goog-api-key).
Spoken Punctuation	Natural voice command conversions ("period" → ., "comma" → ,, "new paragraph" → ↵↵, etc.).
Custom Corrections	Personal find-and-replace word dictionary applied to all transcriptions automatically.
Voice App Launcher	Voice command triggers to launch local applications, scripts, or files (e.g., "open notepad").
Minimal Footprint	Lightweight memory footprint (~80 MB saved by removing redundant WebView2 processes) and bounded history storage.

---

Quick Start

1. Download the latest release installer for your operating system from GitHub Releases.
2. Launch DictateAnywhere. The app lives in your system tray.
3. Focus any text field in any application (Notepad, Word, Browser, IDE, Slack, etc.).
4. Press Ctrl+Alt+D (Windows) or Cmd+Alt+D (macOS), speak naturally, and press it again to inject your text directly at your cursor.

---

Usage & Shortcuts

Action	Control
Toggle Dictation	Press Ctrl+Alt+D (or click Floating Mic Button)
Toggle Text Polish	Press Ctrl+Alt+P
Open Settings	Right-click system tray icon → Settings
View History	Right-click system tray icon → Session History
Move Floating Widget	Click and drag the floating mic button anywhere

Spoken Punctuation Commands

Spoken Phrase	Typed Output
"period" / "full stop"	.
"comma"	,
"question mark"	?
"exclamation mark"	!
"new line" / "line break"	↵
"new paragraph"	↵↵
"open quote" / "close quote"	"
"dash" / "hyphen"	— / -
"ellipsis"	…

---

Cloud STT & API Providers

DictateAnywhere allows seamless selection of your preferred cloud STT provider under Settings → Cloud STT:

1. Google Gemini (Recommended)

• Fast, multimodal transcription powered by Google Gemini 2.0 / 2.5 Flash models.
• Get a free API key from Google AI Studio.
• Paste your key in the Cloud STT tab and toggle Google Gemini as the active provider (authenticated securely via headers).

2. Microsoft Azure Speech

• Enterprise-grade speech recognition via Azure Cognitive Services.
• Create a free resource at azure.microsoft.com.
• Enter your API Key and Region (e.g., eastus) in the Cloud STT tab.

3. Sarvam AI (Indian Languages)

• Specialized for Indian languages and accents (Hindi, Tamil, Telugu, Kannada, Malayalam, Marathi, Bengali, Gujarati).
• Obtain an API key from Sarvam AI.
• Supports both standard REST audio transcription and real-time WebSocket streaming.

---

Local Development & Building

Prerequisites

• Node.js (v18 or higher) & npm
• Python (v3.11 or higher)
• Rust & Cargo — rustup.rs

---

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/RhythmicDias/DictateAnywhere.git
   cd DictateAnywhere

2. Create Python Virtual Environment & Install Dependencies:

   python -m venv .venv
   
   # Windows:
   .\.venv\Scripts\activate
   
   # macOS / Linux:
   source .venv/bin/activate
   
   pip install -e .

3. Install Frontend Dependencies:

   npm install

4. Build the Python Sidecar Binary: The desktop app relies on a bundled sidecar binary for audio capture and inference.

   python scripts/build_sidecar.py

   Note: This generates the platform-specific binary in src-tauri/binaries/.

5. Run in Development Mode:

   # Windows 1-Click Launch:
   .\RUN.bat
   
   # Or via npm:
   npm run tauri dev

---

Building Standalone Installers

To package a production release executable/installer for your current OS:

# 1. Build the updated sidecar executable
python scripts/build_sidecar.py

# 2. Package the native installer (.msi / .exe on Windows, .dmg / .app on macOS)
npm run tauri build

The output installers will be generated in src-tauri/target/release/bundle/.

---

Architecture & Project Structure

DictateAnywhere/
├── src/                         ← React (TypeScript) UI Windows & Overlays
│   ├── windows/Settings/tabs/   ← Modular settings tabs (Engine, Cloud STT, Audio, Text Polish, Hotkeys, Widget, Auto-Correct, App Launcher, Advanced)
│   ├── windows/FloatingWidget/  ← Draggable floating microphone button
│   ├── windows/PreviewOverlay/  ← Real-time transcription preview
│   └── windows/HistoryWindow/   ← Session history viewer
├── src-tauri/                   ← Rust Core (Tauri v2, Win32 SendInput, Keyring, Hotkeys, RFD)
│   ├── binaries/                ← Platform sidecar binaries (whisper_sidecar-*.exe)
│   ├── src/                     ← Rust backend commands & state
│   └── tauri.conf.json          ← Tauri configuration & strict CSP permissions
├── sidecar/                     ← Python Sidecar Core
│   ├── whisper_sidecar.py       ← JSON IPC loop & STT orchestrator
│   └── whisper_sidecar.spec     ← PyInstaller spec with hardened excludes
├── src/dictateanywhere/         ← Lean Python STT & Core Engine
│   ├── audio/                   ← Low-latency audio capture & WebRTC VAD silence stripping
│   ├── transcription/           ← local_engine (FP16 CUDA faster-whisper), gemini, azure, sarvam
│   ├── core/                    ← punctuation normalization, corrections, LLM text polish
│   └── utils/                   ← configuration manager & credentials
├── tests/                       ← Pytest test suite
├── RUN.bat                      ← 1-Click developer launch script
├── pyproject.toml               ← Python packaging & dependencies
└── README.md

---

License

This project is licensed under the MIT License.

---

Acknowledgements

• Tauri v2 — Next-generation cross-platform desktop framework
• faster-whisper — CTranslate2-based Whisper inference
• Google Gemini API — Multimodal AI speech and polish models
• Sarvam AI — Speech technology for Indian languages
• Azure Speech SDK — Cloud speech recognition services