LiveTranslate

Speak in one language, be heard in another — in real time.
100% local. No cloud. No API keys. No data leaves your machine.

Features • Quick Start • Build from Source • Architecture • Contributing

---

What is LiveTranslate?

LiveTranslate captures audio from your microphone, transcribes it with a local AI model, and translates it in real time. The translation can be shown as an on-screen subtitle overlay and/or spoken aloud by a text-to-speech engine.

The main use case is live video calls: you speak in your language, LiveTranslate translates and plays the audio through a virtual audio cable, and the other person on Zoom, Teams, or Meet hears you in their language — with no interpreter and no cloud service involved.

How it works

LiveTranslate works like a simultaneous interpreter — you don't have to stop speaking for the translation to happen:

1. Capture — your microphone audio is captured directly by LiveTranslate.
2. Translate as you speak — Parakeet TDT 0.6B v3 transcribes your speech natively in Rust (GGUF via transcribe.cpp) and MarianMT translates it, with a dedicated model per language direction (EN ↔ ES/FR/DE). While you talk, in-progress translations stream to the subtitle overlay, and each finished sentence is detected and closed on the fly.
3. Speak — every closed sentence is voiced immediately — in your own cloned voice (Pocket TTS) or a standard voice (Piper TTS) — and played through a virtual audio cable: set that cable as your microphone in the video call app and the other person hears you in English, sentence by sentence, while you keep talking.

flowchart LR
    A["🎙️ Microphone"] --> B["Parakeet TDT 0.6B v3\nSpeech → Text\n(streaming, sentence by sentence)"]
    B --> C["MarianMT\nText → Translated Text\n(per language pair)"]
    C --> D["📺 Live Subtitle Overlay"]
    C --> E["Your cloned voice / Piper\nText → Speech"]
    E --> F["🔌 Virtual Cable\n→ Zoom / Teams / Meet"]

Everything runs locally on your machine. No audio or text is ever sent to the cloud.

Virtual audio cable

To route the translated audio into a video call, you need a virtual audio cable driver:

• Windows: VB-Audio Virtual Cable (free) — install it, then in your video call app select "CABLE Input" as the microphone.
• macOS: BlackHole (free) — install it, then select "BlackHole" as the microphone in your call app.

Without the virtual cable, LiveTranslate still works — you see the subtitles and hear the translation through your speakers, but the other party on the call won't hear the translated audio.

---

Features

Simultaneous translation

• Interpreter-style flow — sentences are detected as you speak and voiced while you continue talking; no need to pause
• Live subtitles — the in-progress translation streams to the overlay and refines in real time
• Microphone mode — translate your own speech into another language
• English pass-through — if you speak in English, LiveTranslate detects it and skips translation entirely; the original audio goes through unchanged
• Push-to-talk — hold a hotkey to translate, release to stop
• On-screen subtitles — transparent overlay sits above all windows, click-through, auto-hides
• Audio playback (TTS) — hear the translation spoken aloud (Piper voices)
• Voice cloning — record a short sample of your voice and have the translation spoken back in your own voice instead of a generic one (powered by Pocket TTS)

100% local & private

• All inference runs on your machine
• No cloud services, no API keys, no telemetry
• HuggingFace models downloaded once during setup
• Works completely offline after initial model download

Self-contained setup

• One-click installer bundles a portable Python runtime
• Setup downloads and configures everything: Python, models, voices
• No manual Python installation, no pip install, no environment variables
• Survives app reinstalls (models cached in HuggingFace home directory)
• In-app updates — the app checks for new versions and updates itself with one click (signed releases)

Push-to-talk (PTT) shortcut

The PTT shortcut is user-configurable — click the "Capture" button in settings and press any key combination (e.g. CapsLock, Ctrl+Shift+Space, Ctrl+\). No shortcuts are hardcoded.

In PTT mode: hold the shortcut to record, release to stop and translate.

---

Quick Start

Download the installer

Grab the latest installer from the Releases page:

• Windows: LiveTranslate-windows-x64-setup.exe
• macOS (Apple Silicon, beta): LiveTranslate-macos-aarch64.dmg — unsigned build, see docs/MACOS.md

1. Run the installer
2. Launch LiveTranslate — the setup wizard will download Python and models automatically
3. Select your microphone
4. Choose source and target languages
5. Start speaking — subtitles appear on screen

Requirements:

• Windows 10/11 (64-bit) or macOS 12+ (Apple Silicon)
• ~4 GB free disk space (for Python runtime + models)
• Internet connection on first run (models download once, then fully offline)
• A virtual audio cable (VB-Cable on Windows, BlackHole on macOS) — required to route translated audio into video calls

---

Building from Source

Prerequisites

• Rust (latest stable)
• Node.js 18+
• pnpm
• Tauri CLI

Steps

git clone https://github.com/NBS282/LiveTranslate.git
cd LiveTranslate

pnpm install
pnpm tauri dev      # development mode with hot-reload
pnpm tauri build    # production build

For detailed instructions, see CONTRIBUTING.md.

---

Architecture

graph TD
    subgraph app["Tauri Desktop App"]
        subgraph frontend["Frontend — TypeScript + Vite"]
            MW["Main Window"]
            OW["Subtitle Overlay\nalways-on-top · click-through"]
        end
        subgraph rust["Backend — Rust"]
            AUDIO["Audio Capture\ncpal (WASAPI / CoreAudio)"]
            STT["Native Speech-to-Text\nParakeet-TDT 0.6B v3 GGUF\ntranscribe.cpp — CPU / Metal"]
            SETUP["Setup Wizard"]
            ENGINE_MGR["Engine Manager"]
            CMDS["Tauri Commands & Events"]
        end
    end

    subgraph engine["AI Engine — Python + FastAPI"]
        MT["Translation\nMarianMT — one model per direction\nalternative: Canary 1B Flash (AST)"]
        TTS["Text-to-Speech\nPocket TTS (cloned voice) / Piper"]
    end

    MIC["🎙️ Microphone"] --> AUDIO
    AUDIO --> STT
    STT --> CMDS
    CMDS <-->|"Tauri events"| MW
    ENGINE_MGR -->|"spawns process"| engine
    CMDS -->|"HTTP — localhost\nfinals + streaming partials"| MT
    MT -->|"live subtitle text"| OW
    MT --> TTS
    TTS -->|"audio playback\n→ virtual cable"| MW
    SETUP -->|"first run only"| HF["HuggingFace Hub\nmodels downloaded once"]

Tech Stack

---

Project Status

LiveTranslate is in active development. Current focus areas:

• Windows installer (MSI + NSIS)
• Spanish → English translation pipeline
• Simultaneous translation — sentences voiced while you keep speaking (Parakeet + MarianMT)
• Live streaming subtitles while you talk
• Transparent subtitle overlay
• Push-to-talk
• Multilingual support — EN ↔ ES/FR/DE language pairs
• Language selection UI
• Native speech-to-text in Rust (GGUF Parakeet via transcribe.cpp) — ~3× less RAM than the Python pipeline
• macOS support (Apple Silicon, beta — unsigned build)
• In-app auto-updates (signed releases)
• Voice cloning — instead of a generic TTS voice, the translated audio sounds like your own voice speaking the target language
• Linux support
• System audio mode — translate audio playing on your machine (videos, streams, calls you're listening to)
• Live keyboard output — type the translation as if you were physically typing it, so it appears in any text field on screen (chat, forms, live captions)

---

FAQ

Does it work offline? Yes. After the first setup downloads the models, everything runs locally with no internet connection required.

Does it require a GPU? No. Speech recognition runs natively on CPU (Metal-accelerated on Apple Silicon), and the rest of the pipeline is CPU-friendly too.

What languages are supported? English ↔ Spanish, French, and German, in both directions (default: Spanish → English). You pick the pair in the app.

Is my data private? 100%. Everything runs on your machine. No audio, text, or any data ever leaves your computer.

Does it work with any video call app? Yes. LiveTranslate translates your microphone and plays the result through a virtual audio cable. You then select that cable as your microphone inside the call app. This works with any app that lets you choose an audio input — Zoom, Teams, Google Meet, Discord, and others.

Can I use it to translate audio I'm listening to? Not yet — today LiveTranslate translates your microphone. System audio mode (translating videos, streams, or the other side of a call) is on the roadmap.

---

Contributing

See CONTRIBUTING.md for detailed setup and guidelines.

Quick summary:

1. Fork and clone
2. pnpm install
3. Make your changes
4. Submit a PR with conventional commits

All contributions welcome — bug fixes, features, docs, tests.

---

License

MIT © 2026 Nicolás B. S. — see LICENSE for details.