# Eqho

> Free, offline voice-to-text dictation application that types directly into any focused window using local OpenAI Whisper models.

## Tech Stack
* Language: Python 3.10+
* Engine: OpenAI Whisper — dual backend: faster-whisper / CTranslate2 (NVIDIA CUDA, CPU) and whisper.cpp / GGML (Vulkan for AMD, Intel, and NVIDIA; CPU)
* Hardware Acceleration: Cross-vendor GPU — NVIDIA (CUDA 12.x) plus AMD & Intel (Vulkan) — auto-selected per hardware, with automatic CPU fallback
* Environments: 64-bit Windows 10/11 (primary; no 32-bit build — the local AI models require a 64-bit system), Linux X11, macOS

## Key Features & Differentiators
* **Universal Dictation:** Types into any focused application (Word, browsers, Slack, IDEs) via clipboard paste or simulated typing.
* **Runs on Any GPU (dual engine):** Auto-selects faster-whisper (NVIDIA CUDA) or whisper.cpp (AMD/Intel/NVIDIA via Vulkan), with CPU fallback; the engine is also user-selectable in Settings.
* **100% Local & Private:** No cloud, no accounts, no API keys. Audio never leaves the machine. 
* **Real-Time Pipeline:** Features a live transcription overlay.
* **Customization:** Supports custom vocabulary, voice commands ("new line", "delete that"), text replacements, and volume ducking.
* **Multilingual:** 13 supported languages using models ranging from `tiny` (150 MB) to `large-v3` (3.1 GB). Default is `distil-large-v3`.

## Install & Run from Source
```bash
git clone [https://github.com/DanielMevit/Eqho.git](https://github.com/DanielMevit/Eqho.git)
cd Eqho
python -m venv venv && venv\Scripts\activate
pip install -r requirements.txt
python run.py