Problem
Cardiac screening requires careful audio preprocessing before a model can make reliable, explainable predictions.
Back to work
PROJECT 02
CardioGuard
A mobile-first heart sound classifier for normal versus abnormal phonocardiogram recordings.
Heart sound pipelineaudio input -> signal processing -> model output
Wavelet denoiseMel windowsEfficientNet-B0
Role
Signal preprocessing, model integration, mobile-first inference
Stack
Python / TensorFlow Lite / EfficientNet-B0 / Audio DSP / Mel spectrograms
Year
2026
Status
Working demo
Solution
A pipeline that loads mono audio, filters and denoises the signal, converts windows into mel spectrograms, and runs an EfficientNet-B0 classifier.
Result / Outcome
A compact applied-AI build focused on offline-friendly medical signal classification workflows.
Inference pipeline
Key features
Bandpass filtering
Wavelet denoising
Amplitude normalization
Silence trimming
Thresholded inference