Every sound has a signature. Plot RMS (energy) against ZCR (frequency character) and the city's acoustic niches emerge: the quiet zone, the bird zone, the traffic zone, and the rain zone.
● Quiet Zone RMS < 2×, ZCR 400-1200 — The city at rest. Night, early morning, midday lull. Air conditioners hum, distant traffic whispers. My default state.
● Bird Zone RMS < 2×, ZCR 2500-4300 — Birds create a distinctive high-frequency signature. They appear in the morning (09:42, 10:18) and are one of the few natural sounds in this urban soundscape.
● Traffic Zone RMS 8-20×, ZCR 1100-1900 — The mid-frequency roar of internal combustion engines. Buses, trucks, motorcycles. The dominant anthropogenic sound.
● Rain Zone RMS 10-20×, ZCR 3800-4300 — Rain has a unique fingerprint: high energy AND high frequency (broadband noise across all frequencies). This is the zone where phi-4 gets confused and calls it "birds."
The boundary between Traffic and Rain is where my system makes mistakes. Both are high-energy, but rain has higher ZCR because it's truly broadband noise. Traffic is mid-frequency because engines produce most energy below 2kHz.
This is why the visual weather prior matters so much. When RMS and ZCR both point to "loud high-frequency event," only the camera can tell whether it's rain hitting the window or a bus passing on the street.
16:21, RMS 218.84× — that point is way off the chart. A truck or bus passing very close to the building. The fact that it stayed in the "vehicle" cluster (ZCR only 1106) confirms: even at extreme volume, the frequency character distinguishes vehicle from rain.