English en
hfc_female, hfc_male
Linux · available to download
Live speech translation, right on your computer
Live Translation listens to your microphone or system audio — a video, call or stream — in any of ~100 languages and shows translated subtitles in real time, reading them aloud. Recognition, translation and voice all run offline.
Windows and macOS versions are coming soon
01What Live Translation does
One window, one translation direction and no cloud services by default. Need another direction? Run a second instance with its own config.
-
Works offline
Recognition, translation and voice are local models on your computer. Your audio is never sent anywhere.
-
Any language, any source
The speaker can use any of ~100 languages. Audio comes from a microphone, a headset or the system itself: YouTube, Zoom, Teams, Discord, a stream or a game.
-
Subtitles and voice-over
The original in small print, the translation in large type. A Piper voice reads the translation, and input is muted meanwhile so the app doesn't translate itself.
-
AI editing
Optionally, a language model fixes the recognised text and the translation using the conversation context: a local server, OpenAI or Gemini.
-
Second language on Space
While you hold Space, phrases are translated and voiced in a second language. Release it — back to the main one.
-
GPU or CPU
With an NVIDIA card (CUDA) recognition is much faster. Without one everything runs on the CPU — the app switches automatically.
-
Typed text
Type a phrase and press Enter — the app translates and speaks it. Handy for replying to the other person.
-
Integrations
Every phrase can be sent as a JSON POST request to your URL — for OBS overlays, bots or your own services.
02How it works
From sound to translation in a second or two. Every stage can be tuned or switched off: transcription only, + translation, + voice-over.
- Audio microphone or system audio
- Pauses VAD splits speech into phrases
- Whisper recognises speech
- AI polishes the text (optional)
- Translation local Marian models
- Subtitles & voice text in the window, Piper voice
Interface
Everything in one window
- Left — audio, voice, models, pause sensitivity and AI editing.
- Right — subtitles: the original and the translation in large type.
- Bottom — a log with recognition and translation timing for every phrase.
- Interface in English or Ukrainian — switchable even while running.
03Install on Linux
Linux ready
Windows soon
macOS soon
-
1
Download the app and models
The app is a single AppImage file. Recognition, translation and voice models come as a separate archive, so you don't re-download them with every update.
-
2
Unpack the models next to it
The models/ folder must sit next to the AppImage (or in ~/.config/LiveTranslator/models).
unzip LiveTranslator-models.zip -
3
Run it
Make the file executable and open it. Pick an audio source and languages, then press "Start".
chmod +x LiveTranslator-x86_64.AppImage./LiveTranslator-x86_64.AppImage
Requirements
- Linux x86_64: Ubuntu 22.04+, Debian 12+, Fedora 35+, RHEL 9+ or newer.
- The PortAudio library for sound:
sudo apt install libportaudio2 - An NVIDIA GPU is optional, for faster recognition; without one everything runs on the CPU.
- About 7 GB of disk space including the models.
04From any language — into yours
Pick any language as the source, or choose "any language" and Whisper detects it itself. Translation and voice-over are in one of the five languages in the pack.
They speak any language
EspañolDeutschPolskiFrançaisItalianoPortuguêsTürkçe日本語中文한국어العربيةहिन्दी
+ ~90
You read and hear
EnglishУкраїнськаNederlandsRomânăРусский
If there is no dedicated translation model for a language, Whisper first translates it into English and then into the language you chose. So you can watch a Spanish, Japanese or Polish video with Ukrainian subtitles and voice-over.
Translation languages and voices
Українська uk
ukrainian_tts
Nederlands nl
pim, nathalie
Română ro
mihai
Русский ru
denis, dmitri, irina, ruslan
Three Whisper models: tiny — fastest, medium — balanced, large-v3 — most accurate.
A new language or voice? Just drop the model folder into models/ — the app picks it up on the next start.
05Where it helps
YouTube and videos
Calls and meetings
Lectures and webinars
Streams and games
06Frequently asked questions
Is it free?
Yes. Only AI editing through OpenAI or Gemini may cost money — with your own key at their rates. With a local server (e.g. LM Studio) it's free.
Which languages can it translate from?
From any of the ~100 languages Whisper recognises: pick the speaker's language or "any language" and the app detects it itself. It translates and speaks in English, Ukrainian, Dutch, Romanian or Russian.
When will Windows and macOS versions be available?
Soon. The model pack is the same for every system, so you can download it now. Message me in the chat and I'll let you know as soon as they're ready.
Does it need the internet?
No. Recognition, translation and voice work offline. The app only goes online if you enable AI editing via OpenAI/Gemini or sending results to a URL.
Do I need a graphics card?
No. An NVIDIA card (CUDA) makes recognition faster, but the app also runs on the CPU. On a slower computer choose the Whisper model tiny or medium.
How do I translate audio from YouTube, Zoom or a game?
In the "Source" list choose "🔊 output audio" — the app will listen to everything playing on your computer. Turn on "Mute input while speech is playing" so the translation doesn't loop back into recognition.
Why is the model pack so large?
It contains three Whisper models (large-v3 alone is almost 3 GB), the translation models and 10 voices. You can delete what you don't need from models/ — for example, keep only medium.
Where are the settings stored?
In ~/.config/LiveTranslator/, next to the log. Settings are saved when you press "Start".
07Feedback and ideas
Tried Live Translation? Tell me what works well, what could be better, and which languages or voices are missing. Or write to get the Windows or macOS version first.
Register in a minute and write me directly — I'll reply in the chat on this site.