Charlas - Carla Marcela Florida Román: Clasificación de violencia mediante análisis de audio...

Learn how machine learning and audio analysis can detect violence in real-time through gunshots, screams, and fighting sounds to enhance public safety and emergency response.

Key takeaways
  • Research focuses on developing an audio-based violence detection system using machine learning and audio processing
  • System aims to identify various types of violence-related sounds including:
    • Gunshots
    • Breaking glass
    • Screaming
    • Crying
    • Fighting/disturbances
  • Project faced challenges with limited open datasets for violence-related audio
  • Audio data was collected from multiple sources including:
    • YouTube videos
    • Police recordings
    • Public sound databases like FreeSound
  • Technical implementation includes:
    • 2-second audio segment analysis
    • Conversion of audio to spectrograms
    • Both audio and text analysis components
    • Real-time monitoring capabilities
  • System designed to:
    • Detect violent incidents in real-time
    • Send alerts when violence is detected
    • Help law enforcement respond faster
  • Potential applications include:
    • Urban security systems
    • Domestic violence prevention
    • Public safety monitoring
    • Law enforcement response optimization
  • Project demonstrated successful violence detection in test environments with specific audio markers