What is speech-to-speech translation?
Speech-to-speech translation — sometimes called spoken translation or S2ST — takes audio in one language and returns audio in another. The input is a voice; the output is also a voice. That is the crucial difference from the translation most people know, where you type text and read the result, or a video shows a line of subtitles to scan.
A complete S2ST pipeline does three things: recognizes the spoken words, translates their meaning, and synthesizes a new voice that says them aloud. VoxisLive runs this chain continuously, so the result lands in your ears a few seconds after the original speaker — close enough to follow a conversation, a lecture or a scene as it happens. Your eyes stay on the gameplay, the slides, or the person talking.
Why hearing beats reading
Subtitles work, but they cost you something: pulling your gaze to the bottom of the screen, parsing text, snapping back — many times a minute. In a fast scene you miss the visuals; in a game you miss the moment; in a meeting you stop watching the speaker. Reading is a second task layered on top of watching.
Spoken translation removes that task. The language you understand simply arrives through your speakers while everything you see stays in front of you — the same reason live events use interpreters speaking into an earpiece rather than scrolling a transcript on a wall. Captions still exist if you want them: export a TXT, SRT or VTT transcript after any session. They're a record, not the thing you rely on in the moment.
A native simultaneous interpreter
VoxisLive captures system audio with Windows WASAPI process-loopback — no virtual cable, no extra driver, no bot in your call — and excludes its own output so it never translates the voice it just produced. On Linux it does the same without a driver: VoxisLive creates its own PipeWire virtual sink on demand and routes the app you are translating into it. Install it from the Snap Store with sudo snap install voxis — the Snap Store serves the current published build.
The captured speech goes to a native simultaneous interpreter model: rather than waiting for a sentence to finish, it begins translating while the speaker is still talking and stays a few seconds behind — exactly how a human interpreter works in a conference booth. That short, steady lag is what makes the output feel live instead of stop-and-start.
Two modes
In Video / Game mode translation is one-way: the other voice comes into your language and the original audio is ducked so the spoken translation sits clearly on top. In Meeting mode it is two-way: the other party is translated into your language, and your own speech is translated into theirs and fed into a virtual microphone — with no bot in the participant list.
VoxisLive runs on Windows 10 and 11 and on Linux, and speaks 79 target languages. Get it on the Microsoft Store, or use the paid Bring Your Own Key unlock inside the app with your own Gemini or Qwen key — see GitHub for how it works.