Project Overview

This project aims to build an AI-based system capable of continuously monitoring background sounds in hospital or clinical environments to detect critical medical audio cues. These may include sounds such as coughing, wheezing, crying, alarms, or ambulance sirens. The system will analyze continuous audio streams, extract acoustic features, and classify them using AI models to identify and flag medically significant patterns in real-time.

Objectives

• Preprocessing Pipeline: Apply noise filtering, segmentation, and normalization to prepare input for analysis.
• Acoustic Feature Extraction: Extract Mel-Frequency Cepstral Coefficients (MFCCs), Spectral Roll-off, Chroma Features, and Mel Spectrograms.
• Continuous Audio Monitoring: Develop a real-time system to detect and classify medically relevant background sounds.

Expected Outcome

• Working prototype capable of detecting critical hospital sounds with high accuracy and minimal false alarms.
• Real-time notification and logging of detected medical or environmental audio events.
• Dataset and benchmarks created for future research and model refinement.

How to Apply

We invite researchers, AI practitioners, and developers to submit their proposals for tackling this open project. Proposals should include:

• Literature Review: Discuss prior research in audio event detection, medical sound classification, or related acoustic AI applications.
• Proposed Approach: Outline the AI models, feature extraction techniques, and deployment strategies to be used.
• Dataset Plan: Specify how data will be collected, preprocessed, and augmented—either using open datasets (e.g., ESC-50, UrbanSound8K) or hospital-specific recordings.
• System Design: Describe the overall architecture, including data pipeline, model training, and alert system.
• Innovation: Highlight unique contributions, such as noise-resilient modeling, self-supervised feature learning, or efficient on-device inference.
• Deliverables & Timeline: Provide milestones and expected results at each stage.