Live captions and translation, anywhere.
A lightweight, cross-platform desktop overlay that captions your microphone or system audio in real time — locally or in the cloud.

Windows · macOS · Linux

---

Features

• Live captions from your microphone or system audio (e.g. video calls, media playback)
• Local speech recognition via whisper.cpp — offline, private, free — with selectable model sizes (tiny → medium)
• Cloud speech recognition via the OpenAI Whisper API, if you'd rather trade privacy for accuracy on lower-power machines
• Live translation into another language as captions appear, locally via ONNX Runtime (Helsinki-NLP OPUS-MT) or via the DeepL API
• Fully configurable overlay: font, size, color, background, outline, position, and click-through behavior
• System tray support, autostart on login, and an optional "start captioning automatically" switch
• 8 supported UI languages: English, Spanish, French, German, Polish, Portuguese, Chinese, Japanese

Getting started

Prerequisites

• Node.js + pnpm
• Rust (stable toolchain)
• Platform build tools for Tauri
• cmake and LLVM (libclang) — required by whisper-rs and ort at build time

Development

pnpm install
pnpm tauri dev

Building

pnpm tauri build

Produces platform installers (MSI/NSIS on Windows, DMG on macOS, AppImage/deb on Linux) under src-tauri/target/release/bundle/.

How it works

Audio is captured via cpal (microphone) or WASAPI loopback (system audio on Windows), resampled to 16kHz mono, and streamed through a rolling buffer to the active speech recognition backend. Recognized text is optionally passed through a translation backend and emitted to a transparent, always-on-top overlay window, independent from the main Settings window.

Speech and translation models are downloaded on demand into the app's data directory the first time they're used, and can be managed (downloaded/deleted) from the Models tab.

Tech stack

App framework	Tauri
UI	React + TypeScript
Local speech recognition	whisper.cpp via whisper-rs
Local translation	ONNX Runtime (ort) running Helsinki-NLP OPUS-MT models
Audio capture	cpal, wasapi

Author

Created by RikoDEV