Two-Way AI Translation on Windows: Both Sides, One Overlay

TL;DR: StreamVox translates both sides of a conversation, not just one - so neither person has to read broken subtitles while the other speaks fluently. It works two ways: mirrored across two machines for calls like Zoom and Teams, or on a single machine with Conversation mode for face-to-face talks. Set your languages, turn on the overlay, and both sides read subtitles in their own language in real time.

Most translation tools solve half the problem. You set an input and an output language, the app translates what you say, and the other person still has to struggle through broken English - while trying to reply in a language they barely speak. Half the meaning gets lost either way.

Two-way translation fixes this. Both sides stay in their native language. StreamVox has supported this from day one - it's not a new feature, but it's the one people miss most often when they first open the app, so here's how it actually works.

What Two-Way Translation Means

Traditional translators are one-directional. Input: Japanese. Output: English. Great for watching an anime clip, useless for an actual conversation.

Two-way means simultaneous translation in both directions:

Person A speaks Japanese, Person B reads English subtitles on their screen.
Person B speaks English, Person A reads Japanese subtitles on their screen.

Nobody switches languages. Nobody waits. Nobody says "sorry, can you repeat that more slowly?" It's the closest thing to a real-time interpreter, except the interpreter lives on both laptops and doesn't need a coffee break.

How to Set It Up

StreamVox runs as a transparent floating overlay on top of any app on Windows or Mac. No browser extension, no virtual camera, no rerouting your audio through a sketchy driver. Here's the flow for a typical international call:

1. Pick your languages. Open StreamVox, set your input language (what you'll hear from the other side) and your output language (what you want to read). The other person does the same on their machine, mirrored: their input is your output, and vice versa.

2. Choose the audio source. You can capture everything coming out of your system, or isolate a single app with Per-App Audio Capture. For a Zoom call, point it at Zoom - the overlay will ignore Spotify, notifications, and everything else in the background.

3. Launch the overlay. A compact subtitle window appears on top of your screen. It stays pinned over any application - Zoom, Teams, a game, a browser - without blocking your workflow.

That's the whole setup. Both sides now read subtitles in their own language, in real time - subtitles appear as the words are spoken, no waiting for a paragraph to finish first.

One Computer, Two Languages: Conversation Mode

Everything above assumes two machines - one on each side of the call, with input and output mirrored. That's the right setup for Zoom, Teams, or a Phone Link call, where the other person is on a different device entirely. Sit down across a table from someone, though, and there's no second laptop to mirror.

As of July 2026, StreamVox does something the two-machine flow above can't: Conversation mode turns a single machine into a live two-way interpreter, with no second install and no mirrored setup required. Pick any two of 10 supported languages - English, Spanish, French, German, Italian, Portuguese, Russian, Japanese, Dutch, and Hindi - and StreamVox uses automatic speaker detection to tell who's talking and translate accordingly. Each speaker gets their own color in the transcript, so you can follow who said what without losing the thread.

It's built for face-to-face conversations - a client meeting, an interview, a family dinner where not everyone speaks the same language. The same underlying engine is also what powers the real-time Zoom and Teams translation guide when the other person on a call needs to read your language too, not just you reading theirs.

Where It Actually Works

StreamVox doesn't care which app is making the sound. If your computer can play it, StreamVox can translate it:

Zoom, Teams, Google Meet, Discord - international meetings, remote work, cross-border client calls.
Phone Link (Windows) and Continuity (Mac) - translate mobile calls from an iPhone or Android right on your PC via Phone Link. On a Mac, the iPhone calls you answer on your Mac (via Apple's Continuity) work the same way. Your parents speak Italian, your partner speaks English, everyone stays on one call.
YouTube, Twitch, and other streaming platforms - foreign content with live subtitles, including streams with no official captions.
Games - understand what your Japanese duo partner is yelling in voice chat before you die to it.
Anything else - if it makes noise on your machine, it's fair game.

Languages

StreamVox currently supports 49+ input languages and 49 target languages, covering all major European, Asian, Middle Eastern, and Slavic language pairs. The app interface itself is localized into 12 languages, so you can configure it in whatever you actually think in.

If you want more on how the translation adapts to context - gaming slang vs. business formal vs. movie dialogue - the profile system is covered in the translation modes guide.

Privacy

Audio is processed in real time to generate your subtitles - through trusted third-party AI processors bound by data-processing agreements - and it is never recorded or stored, either by StreamVox or by those providers. No transcripts leave your machine unless you export them yourself. That matters more than it sounds - plenty of "AI translators" quietly retain your audio or transcripts on servers you have no visibility into. If you're on a call about contracts, health, legal matters, or just a private family conversation, you probably don't want that.

Getting Started

StreamVox is a free download on the Microsoft Store for Windows 10 and Windows 11, and it's built for Mac too (macOS 14.2+). One app, four tiers - pick whichever matches how much you actually talk:

Free Starter - 15 minutes every 24 hours, access to all supported languages, system audio and microphone capture.
StreamVox Pro - 40 hours of translation per month, all current and future features, priority email support.
StreamVox Pro+ - 70 hours per month, for people who live in meetings, long streams, or back-to-back movies.
StreamVox Unlimited - no monthly cap, translate all day every day.

Every paid tier gets the same features - the difference is only how many hours you can translate. Search "StreamVox" in the Microsoft Store, or grab it directly from streamvox.pro.

Frequently Asked Questions

What is two-way translation?

It's translation that works in both directions at once, not just one. Each person speaks and reads in their own language, instead of one side getting fluent subtitles while the other struggles through broken text.

Do both people need StreamVox installed?

It depends on the setup. For a call, yes - one copy running on each machine, with input and output languages mirrored. For an in-person conversation, no - Conversation mode runs on a single machine and interprets both sides live.

Which apps does it work with?

Any app that plays audio: Zoom, Teams, Google Meet, Discord, Phone Link and Continuity calls, YouTube, Twitch, and games. StreamVox captures system or per-app audio, so it isn't tied to any one platform.

How many languages does it support?

49+ input languages and 49 target languages for standard two-way translation. Conversation mode, which runs on a single machine, currently covers 10 languages: English, Spanish, French, German, Italian, Portuguese, Russian, Japanese, Dutch, and Hindi.

Is it private?

Yes. Audio is processed in real time to generate subtitles and is never recorded or stored, either by StreamVox or by the AI processors it uses. Nothing about the conversation is retained after the subtitles appear.

Two-way translation sounds like a big deal until you've used it - and then it feels obvious. It's the way these tools should have worked from the beginning. If you've been avoiding international calls because the language barrier makes them painful, try it once. Worst case, you burn 15 free minutes and nothing changes. Best case, you stop needing to translate your own brain before you open your mouth.

AlekGir