Heart Murmur Detection using LSTM-CNN architecture

A deep learning project to detect and classify heart murmurs from heartbeat audio recordings using a Long Short-Term Memory (LSTM) neural network. This repository includes a web application built with Streamlit for a user-friendly interface to process and analyze audio files in real-time.

🚀 Features

• Audio Pattern Analysis: Upload heartbeat sounds (.wav or .mp3) and view their waveforms.
• Deep Learning Model: Uses LSTM-CNN model to classify sounds into 3 categories: Artifact, Murmur, and Normal.
• Real-Time Classification: Automatically extracts Mel-Frequency Cepstral Coefficients (MFCCs) and processes them for inference.
• Hugging Face Integration: Securely downloads and caches the pre-trained Keras model from the Hugging Face Hub automatically.

💻 Tech Stack

• Machine Learning: TensorFlow, Keras, Scikit-learn
• Audio Processing: Librosa, Soundfile
• Web App UI: Streamlit
• Data Handling & Visualization: NumPy, Pandas, Matplotlib
• Model Hosting: Hugging Face Hub

📂 Project Structure

• app.py: Main Streamlit application file.
• audio/: Audio loading and MFCC feature extraction scripts.
• model/: Scripts for downloading and loading the trained model from Hugging Face.
• ui/: Visualization utilities for plotting waveforms in the Streamlit app.
• utils/: Configuration (constants, class mappings) and logging setup.
• Heart_murmur_project.ipynb: Jupyter notebook used for data exploration, preprocessing, model creation, and training.
• requirements.txt: Required Python dependencies.
• check_keras.py: Utility script to verify Keras installation and model loading.

🛠️ Setup & Installation

1. Clone the repository (if not already done):

   git clone <repository_url>
   cd Heart_Murmur_project

2. Create a virtual environment (recommended):

   python -m venv venv
   # On Windows:
   venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install the dependencies: Ensure you have Python 3.8+ installed, then run:

   pip install -r requirements.txt

🎯 Usage

To run the Streamlit web application locally:

streamlit run app.py

This will launch the app in your default web browser (typically at http://localhost:8501). From there, you can upload your heartbeat .wav or .mp3 files and get instant classification results.

📊 Dataset and Resources

• Dataset: The model was trained on the Heartbeat Sounds Dataset from Kaggle.
• Pre-trained Model: The best-performing model is hosted on Hugging Face: ukRani03/Heart_murmur_model.