Google and Watch Duty use Gemini to transcribe noisy wildfire radio feeds
A new AI tool helps a nonprofit wildfire alert app detect fires faster by converting static-heavy radio transmissions into searchable text in under 30 seconds.
The short version
- The tool reduces word error rates on degraded VHF radio audio from 38% down to 22.34%.
- It can transcribe more than 200 hours of daily radio speech from about 400 active feeds.
- The system was designed to handle more than 10,000 feeds at once to scale coverage.
- Volunteer reporter satisfaction with transcription tools rose from 35.3% to 92.3%.
Why it matters
This technology bridges the gap between old analog communication and modern emergency response. By automating the hardest part of monitoring—listening to noisy radio—it allows human volunteers to focus on verifying data and saving lives.
Bridging the analog gap
When a wildfire breaks out, the first signs of the emergency often travel over decades-old analog VHF radio systems. These systems are frequently plagued by heavy static, engine noise, and wind, which can make it nearly impossible for human listeners to understand what is happening. Because of this, the first word of a new fire can be lost in the noise of the radio waves.
Watch Duty, a nonprofit alert app run by first responders, monitors up to 20 of these feeds manually. The challenge for their reporters is cutting through the chaos to find specific information. This manual process can be slow, and in a wildfire scenario, every minute of delay can determine whether residents have enough time to evacuate safely.
To solve this, Watch Duty collaborated with Google to build a dedicated tool powered by Gemini. The goal was to create a system that could automatically wipe out heavy static and translate specialized firefighting jargon into clean, searchable text. This allows the volunteer team to focus on verifying incidents rather than struggling to decipher audio.
Training the model on data
Google engineers and researchers spent six months embedded with Watch Duty to develop this technology. They utilized more than 20,000 hours of past radio recordings to train the Gemini-powered tool. This massive dataset helped the AI learn the specific nuances of firefighting terminology and the unique sounds of radio interference.
The resulting system is designed to handle high volumes of data simultaneously. While it currently processes over 200 hours of daily radio speech from roughly 400 active feeds, it was built to scale to more than 10,000 feeds at once. This scalability ensures that the tool can provide coverage across all 50 states as the organization grows.
The technical success of the tool is measurable. As blog.google reported, the system lowered the word error rate on degraded VHF radio audio from 38% to 22.34%. This improvement significantly lowers the cognitive load on volunteers who must monitor these feeds to keep the public informed.
A real-world test in Washington
The tool faced its first major live test in July 2026 during the Railroad Fire in Washington. The wildfire ignited at night, a time when many people are asleep and the risk of being caught off guard is high. The first 911 dispatch went out over a local fire radio, buried in significant noise and static.
The Gemini-powered transcription tool caught the dispatch and transcribed it in under 30 seconds. This rapid detection gave the Watch Duty team the necessary head start to verify the fire. Because of this speed, they were able to push a Critical Alert, which is a notification that breaks through do-not-disturb settings on mobile devices.
This marked the first instance where Watch Duty issued a Critical Alert during nighttime hours. The successful deployment proved that faster detection can provide residents with the warning they need to evacuate before flames reach their homes. It demonstrated how AI can turn a noisy, old-fashioned radio signal into a modern, life-saving alert.
A multi-year partnership
This project is part of a deep, multi-year collaboration between Google and Watch Duty. It involves funding, cloud infrastructure, and product integration to scale the nonprofit's mission. In 2024, Google.org gave Watch Duty $2 million in funding and $300,000 in Google Cloud credits to help them verify new fires by combining satellite, camera, and radio data.
The partnership also extends to how information is shared with the public. During several major US fires, Google Search included Watch Duty in SOS alerts to appear alongside official emergency instructions. Additionally, Google Cloud raised its messaging limits from 600,000 to 1 million to ensure that push alerts reach Watch Duty's audience of more than 16 million people without delay.
Looking forward, the goal is to turn these models and data into trusted tools for agencies and responders. As part of the AI Collaborative: Wildfires, the teams are working to ensure that when the next fire starts, the warning reaches the right people as quickly as possible. The collaboration aims to provide communities with the time they need to act before a disaster escalates.
What to watch
- How the tool will be scaled to handle 10,000+ feeds across all 50 states.
- Whether other emergency agencies will adopt the Gemini-powered transcription tool for their own dispatch systems.
Try it
Check your phone's settings to ensure that 'Critical Alerts' are enabled for emergency apps so you can receive urgent notifications even if your phone is on silent.
Sources and links
- OfficialAI Collaborative: Wildfires google.org
- Reportingblog.google: the article we sourced
First reported by blog.google. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.