Google Translate 10.32.66 Has a Microphone Setting for Glasses Nobody Can Buy Yet

Every public demonstration Google has given of translation on Android XR has been an argument about the model. At I/O 2025, Shahram Izadi and Nishtha Bhatia stood on stage speaking Farsi and Hindi while prototype glasses put English subtitles in the air between them, and the demo went sideways live before it partially recovered. By MWC 2026 the pitch had moved on to Voice Translate, the Pixel 10 feature that renders a translation in something close to the speaker's own voice, running on prototype glasses Google built itself. The strings that surfaced in the Google Translate app this week concern something less glamorous and considerably harder to get right: which microphone the glasses are supposed to believe.

Android Authority's teardown of Google Translate version 10.32.66 turned up a surface called "Translation settings" and, inside it, an entry reading "Listening area (Glasses only)" with two modes behind it. Front only captures speech from directly ahead and filters out the rest of the room, which is the setting for talking with one person. Surrounding area opens up the full microphone array and translates whoever is speaking nearby. The usual caveat applies, and the publication states it plainly: code found in a shipping APK is work in progress, and a good deal of it never reaches a release.

Set aside whether this particular toggle ships, because the design admission inside it is worth more than the feature. Choosing whose speech to transcribe on a device worn on your face is not a translation problem. It is a beamforming problem, and exposing it as a user-facing setting is Google saying, in the plainest language a settings screen has available, that the platform cannot reliably infer from context whether you mean the person across the table or the room they are sitting in. Anybody who has watched a live translation demo come apart knows the failure is rarely a mistranslated word. It is the system confidently transcribing the wrong voice.

The parenthetical carries the other half of the story. "Glasses only" gates the setting to a form factor Google does not currently sell. The only Android XR device on the market is the Samsung Galaxy XR, out since October 21, 2025 at $1,799, and it is a headset with a microphone array of its own that apparently does not get this control. Frames from Gentle Monster and Warby Parker remain announced rather than dated. So Google is writing form-factor-conditional settings into a phone app for hardware with no ship date, a sequence anyone who shipped through the Android tablet years will recognize on sight: the app arrives first, the device class follows, and for a long stretch in between the same binary behaves differently depending on what it decides it is running on.

For a developer the question is not whether Live Translate gets a listening mode. It is where the listening mode ends up living. If it stays inside Translate as an app setting, then every Android XR application that captures conversational audio, and there are going to be many, rebuilds this decision independently, with its own defaults and its own words for explaining it to a user. If it surfaces instead in the Android XR SDK as a capture mode the platform exposes, applications inherit both the behavior and whatever the wearer has already chosen, and glasses audio starts to look like a platform capability rather than one company's feature. OpenXR will not settle it either way. The standard covers pose, input, spaces and rendering, while audio capture policy sits outside its scope, which leaves the decision exactly where Android has always kept it, in the platform's own APIs.

There is precedent for reading strings this way. In July the same kind of teardown surfaced Spotify code aimed at Android XR glasses before any glasses existed to run it, and the useful conclusion was not that Spotify was launching anything. It was that flagship phone apps are where this platform gets assembled in public, months ahead of the hardware. That remains the most reliable signal available to somebody deciding what to build. Announcements in this ecosystem have a long habit of arriving with no SDK date attached, and the APKs at least come with version numbers.

Worth being precise about what has not happened here. Google has not announced listening modes for Live Translate, has not said the feature is coming to any particular device, and has not confirmed the glasses these strings were written for. What exists is a labeled setting inside a shipping app, plus a run of demos going back to May of last year in which the thing that broke was never the translation itself.

If you are planning to ship something for Android XR glasses in the first half of next year, the practical question this raises is whether to design your own audio capture behavior now or wait for the platform to name one. Waiting is cheaper. It is also how you end up rewriting your input layer in a hurry the week the SDK adds it.