DAVE: Detecting Agitated Vocal Events.

Area of Science:

• Computational Linguistics
• Speech Processing
• Machine Learning

Background:

• Verbal agitations are challenging to detect automatically.
• Existing methods often lack accuracy in identifying diverse vocal events.
• There is a need for robust systems in healthcare and HCI.

Purpose of the Study:

• To introduce DAVE, a novel system for detecting five key verbal agitations.
• To integrate acoustic signal processing with advanced text mining techniques.
• To evaluate DAVE's performance across multiple real-world and controlled datasets.

Main Methods:

• DAVE combines acoustic signal processing with three text mining paradigms.
• Lexical content and acoustic variations are used for detecting specific events.
• Extended word sense disambiguation and sequential pattern mining are employed for cursing and repetitive speech detection.

Main Results:

• DAVE demonstrated significant improvements over baseline methods.
• High accuracy was achieved in detecting all five targeted vocal events.
• Performance was validated across diverse data sources, including dementia patient recordings.

Conclusions:

• DAVE offers a comprehensive solution for detecting critical verbal agitations.
• The system shows broad applicability in healthcare, HCI, and online content monitoring.
• The integrated approach provides accurate and reliable verbal event detection.