Understand this story
The Google Gemini team announced the results of its collaboration with the wildfire-tracking app Watch Duty to improve fire radio transcription. According to the Google Blog, the tool uses Gemini to convert noisy fire radio audio into real-time text, identifying new wildfire dispatch information in under 30 seconds and reducing the word error rate for degraded VHF audio from 38% to 22.34%. The tool currently transcribes more than 200 hours of speech daily across approximately 400 active channels and has been designed and tested to process more than 10,000 channels simultaneously. CNET reported around the same time that the tool was trained using more than 20,000 hours of recordings, and that volunteer satisfaction rose from about 35% to 92%. There have been no subsequent developments to date.
AI-generated from public reports
Latest development · Report
Google Gemini Partners with Watch Duty to Cut Through Fire Radio StaticDevelopments
1 confirmed developments
Google Gemini Team Announces Results of Collaboration with Watch Duty
Source:Google Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 21
CNET
Google Gemini Partners with Watch Duty to Cut Through Fire Radio Static
Google Gemini has partnered with wildfire-tracking app Watch Duty to improve fire radio transcription, transcribing a static-filled 911 call during Washington state's Railroad Fire in under 30 seconds. Trained on more than 20,000 hours of radio recordings, the tool reduced the word error rate by about 15% and raised volunteer satisfaction with the transcription tool from roughly 35% to 92%. It currently analyzes about 200 hours of audio per day from 400 feeds and can scale to more than 10,000 radio feeds.
CNETRelated discussion
CNET joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Nov 20
Google BlogFirst-party
How Watch Duty Uses Gemini to Accelerate Wildfire Alerts
Watch Duty uses Gemini to convert noisy fire-service radio audio into real-time text, identifying new wildfire dispatch information in under 30 seconds. The tool reduced the word error rate for degraded VHF audio from 38% to 22.34%, transcribes more than 200 hours of audio daily across approximately 400 active channels, and has been designed and tested to process more than 10,000 channels concurrently.
AI Trends