Sovereign AI Mic

A completely local, ultra-fast dictation engine featuring a modern web UI, Apple-like frosted glass floating indicators, native Windows text integration, and Google Gemini macro support.

Overview

Sovereign AI Mic lets you transcribe your voice into any application instantly. By holding a hotkey (like Right Alt), your speech is recorded, aggressively transcribed locally using Whisper AI, and seamlessly inserted into your active window natively—bypassing the Windows clipboard altogether.

It bridges the gap between premium local speech-to-text dictation and modern, premium UI/UX design. The application features a decoupled Python backend paired with an Eel/HTML frontend wrapped in a stunning dark-mode interface.

Key Features

• Local Whisper AI Transcription: Choose from Tiny, Base, or Small models. Everything runs offline for dictation.
• Flawless Native Injection: Bypasses clipboard buffering by natively streaming keyboard keystrokes to ensure 100% reliable insertion into every application (even slow browsers, Discord, or IDEs).
• Stunning In-Game UI: Displays an Apple "Frosted Glass" pill indicator with active volumetric equalizers anywhere on your screen.
• Gemini Assistant Integration:
  • Say "Assistant, [question]" to drop a window on-screen and get AI responses.
  • Say "Write a prompt, [idea]" to format complex prompts automatically.
  • Custom Voice Macros to expand short phrases into entire blocks of text (like dropping your email or phone number).
• Zero Privacy Loss: config.json, log.json, and stats.json are stored locally and ignored by source control.

Installation

Prerequisites

• Python 3.10+
• FFMpeg available in system Path (needed for audio transcription)

Setup

1. Clone the repository
2. Install Python dependencies:

   pip install faster-whisper sounddevice numpy pynput eel customtkinter pillow pyperclip

   Note: If you plan on using TTS (Text-to-Speech) modules, you may also need pyttsx3 and edge-tts.
3. Run install.bat which can automatically install missing dependencies (depending on configuration).

Usage

Simply launch the engine using:

run.bat

This will launch a beautiful Web UI Control Center in a Chromium shell. Your floating transcription tool is now active! Just Hold Right Alt (or your mapped hotkey) and speak. Your voice is transcribed and instantly injected into any active window when you release.

Local config (config.json)

• Location: project root (for example, <project-root>/config.json, next to dictation.py).
• The app auto-creates this file on first run and updates it when you save settings in the UI.
• The file is intentionally gitignored, so your Gemini API key and personal macros stay local.

Example structure:

{
  "hotkey_str": "Key.alt_r",
  "hotkey_display": "Right Alt",
  "gemini_api_key": "YOUR_GEMINI_API_KEY",
  "gemini_model": "gemini-2.0-flash",
  "ai_wake_word": "assistant",
  "write_prompt_phrase": "write a prompt",
  "ai_system_prompt": "You are a helpful voice assistant...",
  "voice_macros": {
    "my email": "name@example.com",
    "my phone": "555-0100"
  }
}

Notes:

• Get your Gemini key from aistudio.google.com, then save it in the app's AI page (or set gemini_api_key manually).
• Add any personal/internal info you want to keep private in voice_macros (or related settings) inside config.json.