--- title: "How VoxisLive Works — Driverless Voice Translation (Windows)" description: "Inside the pipeline: WASAPI loopback capture, on-device speech detection, a simultaneous-interpreter AI model and psychoacoustic ducking. See how it runs." url: https://voxislive.com/how-it-works language: "en" source_html: https://voxislive.com/how-it-works site: "Voxis — real-time voice translation" og_image: https://voxislive.com/assets/og.png generated_by: voxislive-markdown-pipeline --- [Home](https://voxislive.com/) · / · How it works HOW IT WORKS # From your speakers to your language in a few seconds. VoxisLive captures system audio directly through WASAPI loopback, detects speech on-device, translates it with a real-time AI model, and speaks the result back — while automatically lowering the original audio. ## Step 1 — Audio capture, without drivers VoxisLive uses **WASAPI loopback** — the same low-level Windows Audio Session API that screen recorders use to capture “what's playing.” It is native Windows capability: no virtual audio cables, no driver installation, no audio routing changes. Capture is zero-latency relative to playback and adds no audible artifacts. Competing tools typically route audio through virtual devices like VB-CABLE, which need driver installs (often admin rights and a reboot) and can conflict with exclusive-mode audio, ASIO drivers or anti-cheat systems. VoxisLive skips that class of problem entirely. ## Step 2 — On-device speech detection The app runs **on-device voice-activity detection (VAD)** to separate speech from silence, background noise and music — locally, on your CPU, with no network round-trip. Only segments identified as human speech proceed to translation, which cuts latency and protects your minute balance. VAD also tracks VoxisLive's own spoken output so it never re-translates its own voice. ## Step 3 — One-pass simultaneous translation Speech segments go to a **multimodal real-time model** that handles recognition, translation and voice synthesis in a single low-latency pass — collapsing the three sequential network calls of a traditional pipeline (speech-to-text → translation → text-to-speech) into one. Like a human interpreter in a booth, it begins translating while the speaker is still talking. ## Step 4 — Spoken playback with ducking The translated voice plays through your output device while two things happen in parallel: **psychoacoustic ducking** lowers the original audio while the translation speaks (mirroring professional simultaneous interpretation), and **latency synchronization** keeps each translation aligned with its speech segment so long sessions never drift. TECHNICAL REFERENCES & CITATIONS ## Architecture & Standards Documentation VoxisLive's engine is built on open standards and enterprise AI specifications. Read the underlying documentation: - **WASAPI Loopback Capture:** Built using [Microsoft Core Audio APIs (WASAPI)](https://learn.microsoft.com/en-us/windows/win32/coreaudio/wasapi) for zero-driver system audio capture. - **Google Gemini Live API:** Real-time WebSocket translation powered by [Google AI Studio Gemini Live API](https://ai.google.dev/docs). - **Alibaba Qwen Realtime API:** Real-time streaming translation powered by [Alibaba Cloud DashScope Qwen Realtime API](https://www.alibabacloud.com/help/en/model-studio/). - **On-Device VAD & Diarization:** Local speaker diarization powered by [sherpa-onnx C++ CPU models](https://k2-fsa.github.io/sherpa/onnx/). ![VoxisLive main window translating a video, showing the live bilingual transcript](https://voxislive.com/assets/shot-app-dark.png?v=1081) ## Watch it live — switching between five languages in one session This screen recording shows a real conference talk translated in real time, with the target language switched live, mid-talk, across Spanish, Italian, Turkish, Chinese and Korean — same source audio, five different spoken outputs, no restart between switches. It's the exact WASAPI loopback pipeline described above: no virtual audio cable, no subtitles, the translation spoken back in a natural voice while the original stays audible underneath. ## How fast is it, really? VoxisLive is **near-simultaneous, not zero-delay**: typically a few seconds behind the original speech, depending on utterance length and network latency. For reference, professional human simultaneous interpreters work two to four seconds behind the speaker — VoxisLive operates in that range or faster. Very short fragments are batched to avoid poor single-word translations. ## Meetings without a bot VoxisLive never joins a call as a participant, never requests host permissions, and never touches the meeting app. It reads the audio already playing to your speakers, which makes it invisible to other participants and identical across Zoom, Teams, Google Meet, Webex and Discord. In two-way mode it also translates your own speech into the meeting language through a virtual microphone. ## What leaves your machine With the managed Store app, audio is proxied to the model and **no audio is retained after the session ends**. On-device VAD separates speech from silence, background noise and music; on paid plans the stream runs continuously while a session is live, and on the free tier only detected speech is sent. GETTING STARTED ## Two modes, set up in under a minute ## Video & Game mode — translate anything playing on your PC Use this for a movie, stream, YouTube video, or game with spoken dialogue. 1. Open VoxisLive and pick **Video & Game** from the left panel. 2. Choose the language you want to hear under **My language**. 3. Press **Start**. Nothing else to configure — VoxisLive listens to whatever is already playing through your speakers and starts speaking the translation within a couple of seconds. There is nothing to install and no separate audio device to pick: VoxisLive captures your PC's system audio directly (see "Audio capture, without drivers" above). ## Meeting mode — be heard and understood on Teams, Zoom or Google Meet Meeting mode is two-way: you hear the other person's speech translated into your language, and — if you want them to hear you translated too — your voice is translated into theirs. 1. Open VoxisLive and pick **Meeting** from the left panel. Set **My language** (what you want to hear) and **Their language** (what the other side should hear from you). 2. Press **Start**. You'll immediately hear the other participant translated — this half needs no extra setup. 3. **To be understood back, one extra step is required in your meeting app, not in VoxisLive:** open Teams' (or Zoom's, or Meet's) own audio settings and set the **microphone** to the virtual cable installed on your PC (named "CABLE Output" or similar) — VoxisLive does not install one, it uses whichever is already there. VoxisLive switches your Windows default microphone automatically, but some meeting apps remember whichever microphone you had selected before and don't pick up the change on their own — so double-check it once, at the start of the call. Skip the last step and Meeting mode still works — you'll hear everyone else translated, you just won't be translated back to them (VoxisLive calls this "listen-only" and says so on screen). No virtual cable installed at all also degrades gracefully to listen-only, never an error. On a WhatsApp call specifically? See the [illustrated WhatsApp call setup guide](https://voxislive.com/use-cases/whatsapp-translation) for exact click-by-click screenshots of both steps. FAQ ## Common questions ### Do I need VB-CABLE or a virtual audio driver? No — not to capture your PC audio. VoxisLive uses WASAPI loopback, a native Windows API available on Windows 10 and 11, so there is nothing to install or route and no new device appears in your audio settings. A virtual microphone is only optional, for sending your translated voice back into a two-way meeting; without one, Meeting mode runs listen-only. ### Does VoxisLive join my meeting as a bot? Never. It captures your own system audio locally, so no third attendee appears in Zoom, Teams or Meet, no permission prompt fires, and no platform integration is needed. ### How much delay should I expect? A few seconds behind the original speech after speech detection and AI processing — the same range as professional human simultaneous interpreters, or faster. ### Does translation quality depend on utterance length? Yes — longer utterances translate better. Very short fragments are batched or deferred to avoid poor single-word translations. ### Can I choose which AI model translates my speech? No manual selection is needed or offered. VoxisLive automatically routes each session to the best-performing engine for your chosen language — currently a mix of Google's Gemini Live and Alibaba's Qwen Realtime, described above — and switches to a backup engine automatically if one becomes unavailable mid-session. The interface does not display which engine handled a given session. ## Keep reading [Translate system audio · VoxisLive captures the whole Windows audio mix and speaks a live translation — browsers, players, games, calls.](https://voxislive.com/translate-system-audio-windows) · [WASAPI Loopback · Translate any Windows system audio in real time without a virtual audio cable.](https://voxislive.com/wasapi-loopback-translation) · [Best translation apps compared · Nine real-time voice translation apps by platform, output and price — including where another tool is the better pick.](https://voxislive.com/best-real-time-voice-translation-apps) · [Use cases · One driverless pipeline for every source: translate game audio, Zoom/Teams/Meet calls, Netflix and YouTube video, live…](https://voxislive.com/use-cases/) Free to start · 10 free minutes every day ## Hear every language, in real time. Runs on Windows 10 and 11 and on Linux — no drivers, no setup ritual, no bot in your call. [Get it on Microsoft Store](https://apps.microsoft.com/store/detail/9P5Z0KVS58RS?cid=DevShareMCLPCB) [Linux · Snap Store](https://snapcraft.io/voxis) [View on GitHub](https://github.com/VoxisLive/voxislive) ## Structured data (JSON-LD) ```json { "@context": "https://schema.org", "@graph": [ { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://voxislive.com/" }, { "@type": "ListItem", "position": 2, "name": "How it works", "item": "https://voxislive.com/how-it-works" } ] }, { "@type": "WebPage", "name": "How VoxisLive Works — Driverless Voice Translation (Windows)", "url": "https://voxislive.com/how-it-works", "description": "Inside the pipeline: WASAPI loopback capture, on-device speech detection, a simultaneous-interpreter AI model and psychoacoustic ducking. See how it runs.", "inLanguage": "en", "dateModified": "2026-07-15", "isPartOf": { "@type": "WebSite", "name": "VoxisLive", "url": "https://voxislive.com/" } }, { "@type": "VideoObject", "name": "Translate Windows System Audio in Real Time — Spoken, Not Subtitles", "description": "VoxisLive translates any audio playing on a Windows PC in real time and speaks the translation out loud in any of 79 languages, switching live between targets — no virtual audio cable required.", "thumbnailUrl": "https://voxislive.com/assets/video-thumb-1280x720.jpg", "uploadDate": "2026-07-15", "duration": "PT5M0S", "contentUrl": "https://www.youtube.com/watch?v=F-wItuPKNYI", "embedUrl": "https://www.youtube.com/embed/F-wItuPKNYI", "publisher": { "@type": "Organization", "name": "VoxisLive", "url": "https://voxislive.com/" } }, { "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "Do I need VB-CABLE or a virtual audio driver?", "acceptedAnswer": { "@type": "Answer", "text": "No — not to capture your PC audio. VoxisLive uses WASAPI loopback, a native Windows API available on Windows 10 and 11, so there is nothing to install or route and no new device appears in your audio settings. A virtual microphone is only optional, for sending your translated voice back into a two-way meeting; without one, Meeting mode runs listen-only." } }, { "@type": "Question", "name": "Does VoxisLive join my meeting as a bot?", "acceptedAnswer": { "@type": "Answer", "text": "Never. It captures your own system audio locally, so no third attendee appears in Zoom, Teams or Meet, no permission prompt fires, and no platform integration is needed." } }, { "@type": "Question", "name": "How much delay should I expect?", "acceptedAnswer": { "@type": "Answer", "text": "A few seconds behind the original speech after speech detection and AI processing — the same range as professional human simultaneous interpreters, or faster." } }, { "@type": "Question", "name": "Does translation quality depend on utterance length?", "acceptedAnswer": { "@type": "Answer", "text": "Yes — longer utterances translate better. Very short fragments are batched or deferred to avoid poor single-word translations." } }, { "@type": "Question", "name": "Can I choose which AI model translates my speech?", "acceptedAnswer": { "@type": "Answer", "text": "No manual selection is needed or offered. VoxisLive automatically routes each session to the best-performing engine for your chosen language, and switches to a backup engine automatically if one becomes unavailable mid-session. The interface does not display which engine handled a given session." } } ] } ] } ```