Respiratory Cycle Segmentation from Audio Signals

🏆 AIOSP 2022 Hackathon – Problem Statement 1

---

Overview

This repository addresses Problem Statement 1 of the AIOSP 2022 Hackathon, which focuses on the segmentation and annotation of respiratory cycles from raw lung sound audio recordings.

Accurate respiratory cycle segmentation is a critical preprocessing step for downstream tasks such as respiratory disease classification and abnormal lung sound detection.

👉 https://iic.ac.in/aiosp-hackathon/

Problem Statement

Segment the respiratory cycles from the given audio files and annotate them accordingly.

Input:

• Raw lung sound recordings in .wav format

Output:

• Individual respiratory cycle segments
• Each segment correctly annotated for further analysis

---

Solution Approach

1. Audio Preprocessing

• Loaded .wav files using audio processing libraries
• Normalized and denoised signals where required
• Converted raw waveforms into suitable representations for segmentation

2. Respiratory Cycle Segmentation

• Identified inhalation–exhalation patterns from continuous audio
• Segmented complete respiratory cycles from each recording
• Ensured temporal consistency and accurate boundary detection

3. Annotation

• Each extracted respiratory cycle was appropriately annotated

• Generated labeled segments suitable for:

  • Disease classification
  • Crackle / wheeze detection

---

Sample Output

Figure 1. Sample Segments of Respiratory Cycles

The figure below illustrates representative respiratory cycles extracted from the original audio recordings, demonstrating clear separation of individual breathing patterns.

---

Implementation & Results

The complete implementation, along with visualizations and segmentation results, is available in the Google Colab notebook below:

Google Colab Notebook https://colab.research.google.com/drive/13EXSjYrrm0M2nSWGJVr13bxcVW75KADM

The notebook includes:

• Audio loading and visualization
• Respiratory cycle segmentation logic
• Annotated segment outputs
• Intermediate plots and analysis

---

Tech Stack

• Programming Language: Python
• Audio Processing: Librosa, SciPy
• Numerical Computing: NumPy
• Visualization: Matplotlib
• Development Environment: Google Colab

---

Outcome

• Successfully segmented respiratory cycles from raw lung sound recordings
• Produced clean, annotated cycle-level data
• Enabled robust downstream machine learning tasks in later problem statements

---

Hackathon Context

• Hackathon: AIOSP 2022
• Achievement: 🥇 Winner
• Domain: Healthcare AI / Audio Signal Processing

---