| title | Heart Sound Classifier |
|---|---|
| emoji | 💓 |
| colorFrom | yellow |
| colorTo | blue |
| sdk | docker |
| pinned | true |
| license | mit |
| short_description | Classify heart sound (phonocardiogram) recordings — normal vs abnormal |
End‑to‑end system that detects cardiac abnormalities from heart sound recordings. It trains a CNN on log‑mel spectrograms and serves predictions via a FastAPI web app with waveform/spectrogram visuals, class probabilities, batch analysis, PDF reports, and session history.
- Live demo (Hugging Face Spaces): https://aryanbaliyan-heart-sound-classifier.hf.space
- Code: https://github.com/AryanGit720/Heartbeat
- 94% record‑level accuracy and 0.986 ROC‑AUC on PhysioNet CinC 2016 test set
- Audio pipeline: 4 kHz resampling, 5 s windows (50% overlap), 64‑mel spectrograms, global normalization
- Training: TensorFlow/Keras, SpecAugment, class weighting, EarlyStopping/ModelCheckpoint
- Web app: drag‑and‑drop uploads (mobile mic capture), waveform/spectrogram, probability bar chart, segment highlights
- Extra features: batch processing (multiple files/ZIP), PDF export, session history with delete
- Productionized: Docker image, health/readiness endpoints, deployed on Hugging Face Spaces (CPU)
- Core: Python 3.11, TensorFlow/Keras, scikit‑learn
- Audio: librosa, soundfile, audioread/ffmpeg
- API/UI: FastAPI, Uvicorn, Jinja2, Chart.js, Matplotlib, ReportLab
- Data/Storage: SQLAlchemy + SQLite (session history)
- Ops: Docker, Hugging Face Spaces, Git LFS (for model)
- Dataset: PhysioNet/CinC Challenge 2016 (normal vs abnormal), >3,000 recordings
- Segment‑level: ~94.6% accuracy on test
- Record‑level (avg probs across segments): ~94.2% accuracy, ROC‑AUC ~0.986
Note: Model predicts two classes: normal, abnormal_other.
- Python 3.11
- ffmpeg on PATH (for MP3/M4A decoding)
- Git LFS (to fetch the model): https://git-lfs.com
- git clone https://github.com/AryanGit720/Heartbeat
- cd Heartbeat
- git lfs install
- git lfs pull # fetches models/tf_heart_sound/best.keras
python -m venv .venv
.venv\Scripts\activate
source .venv/bin/activate
python -m pip install --upgrade pip python -m pip install -r requirements.txt
uvicorn src.app.main:app --host 0.0.0.0 --port 8000
Visit: http://127.0.0.1:8000
- Supported formats: WAV, MP3, M4A/AAC, OGG/WEBM (ffmpeg required). On mobile, you can record directly with your mic.
- Privacy: app stores only filenames, probabilities, and temporary files under /tmp (cleared on restart). No PHI.
- Medical disclaimer: for education/research only; not a medical device; not for diagnosis or treatment.
- Dataset: PhysioNet/CinC Challenge 2016 — https://physionet.org/content/challenge-2016/1.0.0/
- Libraries: TensorFlow/Keras, librosa, FastAPI, Uvicorn, SQLAlchemy, Matplotlib, ReportLab
- Deployment: Hugging Face Spaces