Abstract / Summary
Accurate detection of valve problems using Phonocardiogram (PCG) sound signals is important for medical action. But listening with a stethoscope by hand is very subjective can have mistakes. Depends heavily on the doctors experience. This paper introduces an end-to-end automated system to classify PCG heart sounds into four types: Normal Aortic Stenosis, Mitral Regurgitation and Mitral Stenosis. The system turns time-domain audio signals into 2D Mel-Spectrograms using Short-Time Fourier Transform (STFT). Key acoustic features—Spectral Centroid, Root Mean Square (RMS) Energy and Zero Crossing Rate (ZCR)—are extracted to catch signs of turbulent murmurs and frequency changes. These normalized features are fed into a Machine Learning Engine using a Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel to create decision lines. To make this AI tool useful in clinics we built a browser-based web interface using Gradio. This interface shows live 2D Mel-Spectrograms gives alerts when something abnormal is found and includes an HTML5 Base64-encoded audio player to fix playback issues across different browsers. On test data the model achieved 100% accuracy, precision, recall and F1-score for all categories. The confusion matrix was perfectly diagonal meaning no misclassifications occurred. The lightweight design makes this system a reliable screening tool for heart health checks in resource clinical environments.