Students translating the lecture in their heads
(01) The problem
A teacher speaks in English, with Hindi mixed into the same sentence. That is how people really talk. Some students in the room think in Tamil, or Odia, or Bengali. They turn each sentence into their own words, and the class has moved on.
(02) What I built
One microphone feeds software I wrote and run. It turns what he says into English text, even when Hindi and English share one sentence. Then each finished line goes out in the language a student picked. The teacher is still on that point. Twenty-two languages are set up. Ten run at once.
Both pictures are drawings of the real thing, not photographs of it. The shape is what runs; the sentence being taught in them is made up.
(03) What went wrong
Four minutes in, the captions stopped. He kept talking.
Nothing crashed. So I read the loudness numbers it prints anyway. The room's hum had pushed the bar for speech above the teacher. It had decided he was background noise.
The bar is now capped at half of his loudest recent moment. Noise can raise it, never above the person talking. A test called no choke fails if unbroken speech stops reaching the screen.
(04) What happened
A whole hour now reaches the phones, in up to ten languages at once. Quiet is never a mystery.