Google says its AI flags wildfire dispatches for Watch Duty in under 30 seconds
The Gemini-powered tool turns noisy emergency radio into searchable text. Google’s new results describe faster detection, not automatic public warnings.
Loading page…
The Gemini-powered tool turns noisy emergency radio into searchable text. Google’s new results describe faster detection, not automatic public warnings.
Listen to this story
Google’s Gemini-powered system is now processing Watch Duty’s live emergency-radio feeds, helping reporters notice dispatches they might otherwise miss while they are monitoring other channels. Google says transcription errors on degraded VHF audio fell from 38% to 22.34%, but Watch Duty reporters still assess and verify incidents before alerts go out. The deployment shows how AI can expand real-time monitoring coverage without taking over a consequential public-safety decision; its reported capacity of 10,000 feeds is tested capacity, not current usage.
The system transcribes more than 200 hours of radio speech daily across about 400 active feeds.
Google says engineers and researchers spent six months working with Watch Duty and used more than 20,000 hours of historical recordings.
Google reports the tool flags dispatches in under 30 seconds, including where no automated dispatch detector is available.
A wildfire dispatch buried in radio static can now reach Watch Duty’s reporters in under 30 seconds, Google said in results published November 20. Its Gemini-powered transcription tool helps surface incidents for people to verify—not send public warnings on its own. The deployment pairs faster radio monitoring with a human check before an alert.
Watch Duty’s bottleneck starts with the source of early fire information. Emergency crews use analog VHF radio, where wind, engine noise and static compete with fast-moving firefighting jargon. Google says individual Watch Duty reporters manually monitor up to 20 feeds. Listening closely to one channel does not solve the problem of following many at once.
The new tool converts that audio into clean, searchable text, accounting for heavy static and specialized terminology. Google says engineers and researchers worked full-time inside Watch Duty for six months through a Google.org Fellowship. They used more than 20,000 hours of historical radio recordings to build the system.
That work follows the fellowship’s April 22 announcement. Watch Duty then described a project to monitor live emergency radio and surface important signals while supporting, rather than replacing, human decisions. Its trained reporters were already following hundreds of public radio streams. The November update supplies operational results for that previously announced effort.
Google’s accuracy figures show improvement, not error-free listening: the word error rate on degraded radio audio fell from 38% to 22.34%. That is a measure of transcription quality, not a claim that every fire is detected or every incident detail is correct. Watch Duty’s stated workflow still leaves assessment and verification to people.
The coverage benefit is separate from that accuracy improvement. Google says the system signals reporters in less than 30 seconds, including in places without another automated dispatch detector. It can therefore bring a dispatch to someone’s attention when no one was listening at that moment, rather than only make an already-monitored channel easier to understand.
Current use and tested capacity also differ. According to Google, the tool transcribes more than 200 hours of radio speech daily from about 400 active feeds. It was designed and tested to handle more than 10,000 feeds simultaneously. The larger number describes tested capacity, not the number of feeds currently in use.
Google points to Washington’s Railroad Fire in July 2026 as a live example. The fire began at night, and the tool transcribed and flagged the first 911 dispatch from noisy local fire radio in under 30 seconds. Watch Duty’s team then verified the fire and issued its first-ever overnight Critical Alert.
Critical Alerts can break through a phone’s do-not-disturb setting in urgent situations. Google says the warning helped residents evacuate before flames reached homes. The sub-30-second figure covers the tool’s transcription and flagging of the dispatch; verification and the public warning followed. For Watch Duty’s reporters, the change was an earlier signal to investigate while the community slept.
Loading discussion...
Join the conversation
Explain whether human verification would change your answer.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.