EchoSub is a desktop application that provides live, fully local subtitle translation for audio played through a computer's speakers. It detects the spoken language and displays translated captions in a language chosen by the user. All speech recognition and translation processes run locally on the user's GPU or CPU, ensuring that no audio data leaves the machine. The application appears as a transparent, borderless overlay at the bottom of the screen. This overlay remains invisible except for the captions. Moving the mouse over the overlay reveals a thin strip with a drag handle, a settings button, and a close button. The overlay can be resized from any edge or corner. Typical use cases include watching foreign language videos, listening to music or podcasts, and understanding livestreams or calls. EchoSub is built with Tauri 2 and Rust. It supports AMD, NVIDIA, and Intel GPUs through vendor-neutral backends, with CPU fallback. The application runs entirely as local processes communicating over 127.0.0.1. The only network traffic required is a one-time download of the AI models. On first launch, the application prompts the user to download the models, which are approximately 3.1 GB for the default balanced preset. The download is resumable. Disk usage varies by preset: fast uses about 2.8 GB, balanced uses about 3.1 GB, and quality uses about 7.9 GB. On Windows, the application requires the Microsoft Edge WebView2 runtime and an installed graphics driver. The Visual C++ runtime is included in the package. Users can manage model downloads through the settings menu or via a provided command-line script for offline preparation.