When every room hears you, which one answers?
Testing whether wake-word loudness can pick the nearest device in a multi-room setup built from commodity speakerphones.
Automatic gain control flattens distance so much that beyond a few feet the "nearest" device is decided by which way you are facing, not how close you are.
The question
In a house with a microphone in every room, one wake word often reaches several devices at once. Exactly one should reply, and it should be the right one. Is the loudest detection a good enough signal for that?
Setup
Matched pairs of identical ~$35 USB speakerphones on Raspberry Pi-class boards. Controlled distance sweeps, with the units swapped between positions to separate effects caused by the device from effects caused by its location. The whole program cost under $100.
Findings
- Devices that hear the same wake word trigger within 6–150 ms of each other, comfortably inside the natural ~400 ms pause after a wake word. Arbitration has time to happen.
- The speakerphones' automatic gain control compresses the distance signal to roughly half of inverse-square at a 2× distance ratio. Beyond about four feet the difference is negligible.
- Which way the speaker is facing shifts the level by 1.5–3.75 dB, about the same as doubling the distance.
Takeaway
On matched hardware, wake-word loudness works as a close-range tiebreaker. At normal conversational distances it can't tell which device is nearest. Reliable "nearest device" selection needs presence sensing beyond audio.