LiveSubs

Real-time AI subtitle overlay for Windows. LiveSubs listens to your system audio (whatever is playing through your speakers or headphones — a game, a movie, a YouTube video), transcribes the English speech live using AI, translates it to Polish, and displays it as an on-screen subtitle overlay — just like movie subtitles, generated on the fly.

It works with any application that plays audio through Windows — not just one specific game. As long as the sound goes through your system's audio output, LiveSubs can subtitle it.

How it works

┌─────────────┐    ┌──────────────┐    ┌─────────────┐    ┌─────────────┐
│ 1. AUDIO    │───▶│ 2. WHISPER   │───▶│ 3. DEEPL    │───▶│ 4. OVERLAY  │
│  capture    │    │ transcription│    │ translation │    │ on screen   │
│ (WASAPI)    │    │  (EN text)   │    │  (EN → PL)  │    │  (PyQt6)    │
└─────────────┘    └──────────────┘    └─────────────┘    └─────────────┘

Audio is captured in small chunks via WASAPI loopback, transcribed locally with faster-whisper, translated through the DeepL API, and rendered on a frameless, always-on-top, click-through-friendly overlay window. Capture, transcription and UI each run on their own thread so audio is never dropped while a previous chunk is still being processed.

Features

• Live EN → PL subtitles for any system audio source (games, browser video, streaming apps)
• Frameless, transparent overlay — always on top, doesn't show up in the taskbar/Alt+Tab, draggable to any position on screen
• System tray icon — show/hide the control panel or subtitles, update your DeepL key, or quit, without hunting for a close button on the overlay
• Control panel to:
  • pick which audio device to capture (speakers vs. headphones, auto-detects available loopback devices)
  • adjust subtitle font size and background opacity live
  • show/hide the overlay
  • set or change your DeepL API key at any time
• Works without a DeepL key too — skip the setup and LiveSubs still shows English subtitles from Whisper
• Silence filtering so the app doesn't try to transcribe/translate quiet or empty audio chunks
• Packaged as a single Windows installer — no Python installation required for end users

Tech stack

Stage	Library	Why
Audio capture	sounddevice / pyaudiowpatch (WASAPI loopback)	Captures whatever is playing on your speakers/headphones
Transcription	faster-whisper	~4x faster than stock Whisper, runs fully offline/locally
Translation	deepl (API)	Best-in-class EN → PL translation quality, free tier available
UI / overlay	PyQt6	Transparent, always-on-top, frameless window + native tray icon
Packaging	PyInstaller + Inno Setup	Single .exe and a proper Windows installer

Installation (end users)

1. Download the latest installer from the Releases page (LiveSubs-Setup.exe).
2. Run it — it installs LiveSubs with Desktop and Start Menu shortcuts, no Python required.
3. On first launch, LiveSubs will ask for a free DeepL API key to enable Polish translation. You can skip this and get English-only subtitles, and add a key later from the tray/control panel.
4. Play anything with English audio through your speakers or headphones — subtitles appear at the bottom of your screen.

Running from source (developers)

Requires Python 3.11+ on Windows (audio capture needs WASAPI, so this only works fully on Windows — see Limitations).

git clone https://github.com/achelminska/live-subs.git
cd live-subs
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
copy .env.example .env
# edit .env and add your DEEPL_API_KEY
python main.py

To build your own installer, see the scripts in tools/ (build.ps1 for the PyInstaller .exe, installer.iss for the Inno Setup installer, install.ps1 to install locally with shortcuts).

Known limitations

• Windows only. Audio capture relies on WASAPI loopback, which doesn't exist on macOS/Linux. The AI/translation/UI modules (transcribe.py, translate.py, overlay.py) can be developed and tested cross-platform, but live audio capture and .exe builds require a real Windows PC — virtual machines with emulated audio hardware are not reliable for this.
• Requires internet for translation. Transcription runs fully offline, but translation goes through the DeepL API. Without internet, LiveSubs falls back to showing the original English text instead of crashing.
• ~1–1.5s latency between speech and subtitle, inherent to the chunk-based approach (the app has to hear a full chunk before it can transcribe it).
• DeepL free tier limit: 500,000 characters/month. Check usage at deepl.com/pro-account/usage.

License

MIT — see LICENSE.