Skip to guide
StreamTranslate

Bilingual audio routing

Plan captions for a two-person bilingual call

A two-person call is harder to caption than a single microphone because each speaker can need a different input language and a different output. If your microphone and the remote caller arrive as one mixed signal, a captioning tool cannot be assumed to separate them reliably. Draw the audio paths before choosing the display.

By StreamTranslate · Updated

Keep your microphone and the remote call audio separate, choose which audience needs which translation, and test each direction independently. StreamTranslate’s standard single-source workflow should not be treated as a released, automatic two-way call translator or native Zoom integration.

Define the two translation directions

Write down a concrete example: you speak English and want the remote participant to read Japanese; they speak Japanese and you want English. Those are two independent tasks. Captions for people watching an OBS broadcast are a third task, because the audience may need both speakers identified on screen.

Choose whether the call platform’s own translated captions meet the participants’ needs before building a broadcast overlay. An overlay in your outgoing video is not automatically readable inside the other participant’s meeting interface, and it does not send translated speech back to them.

Map the audio before configuring software

Separate the intended speech paths
InputWhat it should containWhat to exclude
Your microphoneYour voiceThe call playing through speakers
Call audioThe remote participantYour own monitored microphone
Broadcast mixThe sound viewers should hearDuplicate delayed copies

Use headphones. A loopback or virtual audio device may be needed for some desktop call apps; that routing is platform-specific and must be established before captioning. Never assume a Browser Source captures the call.

Use a supported single-source path first

If your immediate goal is to translate the guest for your broadcast audience, route that guest’s speech into an identifiable OBS audio source and test the StreamTranslate filter on that source. Set the guest’s spoken language and your chosen subtitle language in Control.

If the material is available in a supported browser tab, Elite Watch Party can capture that tab’s audio. It does not automatically distinguish every participant within the tab. Start with one direction and one clear speaker so you can identify routing mistakes before experimenting with additional channels.

Check a two-way design with separate tests

  1. Ask the guest to speak while you remain silent; check which input meter and caption output respond.
  2. Ask the guest to pause while you speak; verify that the guest-only caption path remains quiet.
  3. Try a normal turn-taking exchange with names, numbers and a short question.
  4. Check the audience display and the participant experience separately.
  5. Test mute, headphones and reconnect behavior before an important call.

A setup that produces readable words is not yet proof of correct speaker attribution. If both paths react to one person, fix the duplicate routing before adjusting translation settings.

Choose an appropriate fallback and scope

A specialist two-channel tool can be appropriate when simultaneous bilingual participant captions are essential. The linked open-source example illustrates a two-input architecture; its dependencies, language support and operating-system requirements belong to that project, not to StreamTranslate.

Do not book a meeting on the assumption that a single mixed caption feed will provide independent speaker names, two-way translation or a transcript. Agree on turn-taking and a human clarification method. For a standard broadcast, the simpler supported path is one deliberate speech input with a clearly labeled translated overlay.

Quick answers

Can I mix both voices into one input?

You can route a mix, but that does not establish reliable language separation or speaker attribution. Independent translation directions need a design and tests that preserve the relevant inputs.

Does an OBS subtitle overlay translate what the guest hears?

No. It places text in the video composition. Translated audio or meeting-native captions require their own supported integration.

Is named two-speaker mode promised here?

No. This is an audio-planning guide. It does not announce a public named-speaker room feature or certify a particular two-microphone installation.

Continue with your setup

Sources and further reading

Reviewed October 28, 2026. Follow the linked provider instructions for current availability and setup requirements.