Description:
Each Coast Guard Command Center has a human listening to radio distress frequencies - namely listening to static for 12 hours per day. This is a monotonous, but a critical duty. Because of the nature of these calls, and the numerous hoax calls, it is often difficult for a human to glean the criticality of these calls and get all the required information from them. The challenge is a watchstander may be less attentive after spending 12 hours throughout the day listening for distress activities.
An Artificial Intelligence (AI) system could consume the audio, analyze the audio in real time, learn to detect a real call from a hoax, and then send the information to a watchstander. This system could alleviate hundreds of watchstanding positions therefore, making Search and Rescue (SAR) distress monitoring much more successful. An Artificial Intelligence/Machine Learning (AI/ML) system could listen to thousands of historical calls to learn proper detection techniques and utilize voice inflection technology to greatly improve existing capabilities. The AI/ML system would need to cue watchstanders and potentially initiate processes such as initiating a case file within the Marine Information for Safety and Law Enforcement (MISLE) database. Connection to standard Coast Guard networks such as CG1View may be necessary for watchstander cuing through an existing platform. Potential use cases should be evaluated to determine future opportunities and connectivity requirements.
Voice and acoustic forensics applications already exist and have been used to evaluate known hoax maritime distress calls. This capability was employed after repeat hoax calls had already become a problem in a specific location. In this case, a specific individual had made several hoax calls costing the U.S. Coast Guard thousands of dollars in response operations. Recorded audio files were analyzed to identify acoustic characteristics that helped prosecute this hoax call case and others like it. Rather than waiting to use this technology on the back end as in this example, proactively employing voice forensics in radio frequency monitoring technology could cue watchstanders to anomalies that could possibly be associated with hoax calls. Additional potential use cases such as the acoustic detection of intoxicated mariners and subsequent cuing to watchstanders should be evaluated. Acoustic forensics and AI technologies should be evaluated for their ability to account for and differentiate between other variables that may be encountered such as different regional dialects and accents.
For the proposed solution, an AI/ML application would monitor radio frequencies, relieving watchstanders from the tedious task of listening for hours where cost and time savings could be realized by employing Coast Guard personnel with other mission critical tasks. For initial capability exploration, the AI/ML should monitor VHF distress frequencies between 4 and 10 MHz. VHF Channel 16 is the single channel within this range, but there are several simultaneous sources from different Rescue 21 towers. Building upon the initial capability, AI/ML monitoring could be expanded into the UHF or other Coast Guard VHF working channels.
