Early wildfire reports can be buried in fire-radio traffic distorted by wind, engines and static. Google and nonprofit alert service Watch Duty have built a Gemini-powered tool to convert that audio into searchable text and flag possible dispatches for human review. Google says the system identified a 911 dispatch in less than 30 seconds during a nighttime fire in Washington state.
Key points
- Turning radio audio into leads. Firefighters in many areas still use analog VHF radios, where background noise and specialized terminology can make urgent messages difficult to decipher. Watch Duty reporters have monitored as many as 20 feeds manually. The new tool was developed using more than 20,000 hours of historical radio recordings, with the aim of reducing listening workload and making spoken information easier to search.
- AI flags; people verify. The system sends Watch Duty a signal when it detects a possible dispatch. Reporters then check the information and decide whether an alert should be issued. It is therefore an aid to detection and triage, not an autonomous system that confirms fires and directly alerts the public.
- Performance figures from Google. The company says the tool lowered the word error rate on degraded VHF audio from 38% to 22.34%. It transcribes more than 200 hours of radio speech per day across about 400 active feeds, and was designed and tested to handle more than 10,000 feeds. Volunteer reporter satisfaction with radio transcription tools reportedly rose from 35.3% to 92.3%. The supplied material does not include an independent evaluation of these figures.
What the nighttime incident shows
Google says the tool transcribed and flagged the first 911 dispatch in under 30 seconds when Washington’s Railroad Fire broke out at night in July 2026. That gave Watch Duty staff time to verify the incident and send a Critical Alert, which can break through a phone’s do-not-disturb setting. The organization said it was the first time it had issued such an alert overnight, and that residents received an evacuation warning before flames reached homes. The example illustrates a shorter path from detection to verification; it does not mean the AI replaced dispatchers, fire agencies or the people responsible for publishing alerts.
Significance and limits
The project applies speech recognition at the front of an emergency-information chain. When staff must cover many radio feeds, automated transcription can help them locate potentially important reports faster. But operational value depends on more than recognition accuracy: teams still need to validate leads, coordinate with communities and deliver timely warnings. Google also describes funding, cloud credits and engineering support for Watch Duty. The material does not report false-alarm rates or performance across different regions, accents and radio equipment. The figures and case study should therefore be understood as results reported by the partners, with accuracy, coverage gaps and human review remaining important considerations as deployment expands.
Source: Web Search
Comments
Checking sign-in status...
Loading comments...