U.S. Coast Guard personnel monitoring radio transmissions of vessels at sea must be able to identify distress calls despite interference, degraded conditions, and the stressed speech of those in emergency situations. However, not all distress calls are genuine. Hoax calls can be intended as malicious pranks, due to intoxication, or to distract Coast Guard assets from other tasks such as drug interdiction, and can be difficult to identify. Because responding to false emergencies is costly and can distract from attending real emergencies and on-going missions, researchers have explored automated techniques for hoax detection. Detection of deceptive speech is an active research area, but approaches to-date suffer limitations including the quality of available distress call datasets, the need for large quantities of speech to analyze linguistic structure, degraded audio, and the difficulty of assessing psychological states from voice recordings. Eduworks Corporation proposes to overcome these limitations with Distress Evaluation: Situational Cueing, Alerting and Monitoring (DE-SCAM). We will develop a proof-of-concept DE-SCAM prototype applying acoustic forensics based on our work in Artificial Intelligence/Machine Learning (AI/ML) - based speech processing. DE-SCAM will demonstrate continuous monitoring of radio voice transmissions, analyze the communication for potential hoaxes or fake distress calls, and provide analytic capabilities to profile the voice of the speaker with respect to characteristics such as possible intoxication, accent, and matches against previous hoax profiles. This work will benefit tactical response teams that rely on radio communications for timely incident response, and has potential to benefit telephonic voice communications as well.
