Where a bionic ear stops working, and what we do about it.

Four people at a table. Two side conversations. Someone at the next table laughs. Your dinner companion asks a question and you get the prosody, the timing and the fact that it was a question, but not the words. You say "sorry?" for the third time and then stop asking, and spend the rest of the meal nodding at the right intervals.

Leo: the array steers toward whoever you are facing, suppresses the flanking tables, and the caption line carries the words the beamformer could not fully clean up.

A student follows the lecture adequately. Then the class splits into groups of five, thirty other students start talking, and the room becomes unusable. They spend the session copying a neighbour's notes and reconstructing the discussion afterwards.

Leo: in lecture mode the array locks the podium and holds it through head movement. In group work, captions with speaker attribution mean a glance identifies who is talking without having to find them first.

Six people, one speakerphone, two of them talking over each other. You catch "…can you own that by Thursday?" and have no idea whether it was aimed at you.

Leo: this is the case the spatial map was built for. Each caption line is tagged to the direction it arrived from, so the display can say who, not just what, the single highest-value output the visual layer produces.

Three conversations at one table. A grandchild says something quietly, everyone laughs, and you laugh a beat late because you are reading the room rather than hearing it. The joke is gone before anyone could repeat it.

Leo: captions hold the last few lines, so a beat-late glance recovers what was said instead of losing it. Attribution means you know which grandchild.

A pharmacist behind glass, masked, explaining a dosing change. You hear a muffled sentence containing a number. You are not certain whether it was "one to two" or "one or two", and the difference matters.

Leo: the array steers hard at the counter and suppresses the room. The caption carries the number, which is the part you cannot afford to guess.

A platform change is announced. Everyone around you picks up their bags and moves. You watched them react before you understood the announcement, and now you are following a crowd without knowing where it is going.

Leo: the caption path matters far more than the acoustic path here. Recognition is run against the diffuse field and rendered as text, because text survives reverberation in a way that audio does not.

Subtitles are on. Someone asks you a question during a scene. You now have to choose between the subtitle and the question, and you lose whichever you did not pick.

Leo: the room and the screen are different sources in the spatial map, so the person beside you can be attributed and captioned separately from the television.

Your passenger gives directions at 70 km/h with the windows cracked. You either take your eyes off the road to lipread, or you guess at the turn.

Leo: the array steers to the passenger seat and suppresses road noise, which is broadband and diffuse and therefore exactly what a beamformer is good at. Captions are suppressed while driving.

You fingerspell your name for the cup. The frame speaks it. The barista writes it down correctly on the first attempt, which is not something a phone screen turned around at arm's length reliably achieves.

Leo: fingerspelling recognition plus a phrase set covering common transactional exchanges. You confirm before anything is spoken aloud.

The booked interpreter does not show. The appointment happens anyway, through a combination of writing, gesture and a family member who signs a little and is now hearing your medical history.

Leo: can carry the inbound side well, captioning the clinician is squarely within what the visual layer does. The outbound side is where we are cautious.

Your update is three sentences about a build failure. Typing it takes ninety seconds while eight people watch, so you type one sentence instead and the detail is lost.

Leo: a per-team vocabulary makes this tractable in a way open-domain translation is not, the words recur every day and the structure barely changes.

You choose seats by which ear faces the room, rearrange yourself at every table, and walk on a particular side of whoever you are with. It becomes automatic and it never stops being a calculation.

Leo: the array spans the whole head, so capsules on the unimplanted side still contribute. Geometry does what a single behind-the-ear microphone cannot.

Someone with well-fitted aids manages one-to-one conversation comfortably and gives up on group dinners entirely, a pattern that shows up long before anyone would consider an implant.

Leo: the acoustic front end is indifferent to what is downstream. Modern aids expose accessory streaming much like implant processors do.

Implanted eighteen months ago. Phone calls are manageable, one-to-one is fine, and the weekly team lunch is still a write-off. Two years into an ASL class and not yet confident enough to use it with strangers.

Leo: the inbound channel is immediately useful. The outbound channel becomes useful as signing improves, the same device covers both ends of that transition.