A language selector looks simple until a conversation stops using just one language.

People can begin a meeting in one language, switch when someone joins, and switch back a few minutes later. It is not every meeting, but it happens often enough that asking people to choose one fixed language can interrupt the conversation.

We expanded Speechara.AI in two ways.

First, automatic language detection can identify the language being spoken without requiring the user to choose it before every meeting.

While testing Speechara.AI on mobile, we ran into a limitation that was not obvious at first.

On desktop, Speechara.AI can capture system audio from Zoom, Google Meet, Webex, and other meeting apps, then transcribe and translate the conversation in real time.

Mobile browsers work differently. On both iOS and Android, they can usually access the microphone, but they cannot reliably capture system audio from an online meeting playing on the same phone.

Using the microphone as a workaround does not solve the problem. Mobile operating systems intentionally reduce or suppress speaker audio during calls to prevent echo.

We thought a compact mode was enough for an AI assistant used during meetings.

In Speechara.AI, minimizing the main workspace leaves a small always-on-top view so the user can continue following the live transcript and translation. It seemed like a reasonable balance between visibility and convenience.

Then a tester described a situation we had underestimated: they sometimes share their entire laptop screen with other people in the room and do not want their personal AI workspace to remain visible.

Their point was simple: smaller is not the same as invisible.