SIRAJA

An academic heart-sound classification app using 52 MFCC features and a CNN-LSTM model to predict Normal, Murmur or Artifact.

Source code

(Year): 2025
(Role): Academic Team Contributor
(Stack)
(Type): Academic Team Project / Audio Classification

SIRAJA cover: heart-sound prediction page in a laptop mockup, with a tablet showing waveform and spectrogram and a card of the Normal, Murmur and Artifact classes, on a dark blue background.

SIRAJA home category cards for Normal, Murmur and Artifact; performance tiles excluded.

(Problem)

Explore how recorded heart sounds can be processed and classified in an interactive academic prototype.

(Approach)

Our team used 52 MFCC features with a CNN-LSTM model and built a Streamlit interface for audio upload, class predictions, waveforms and spectrograms.

(Result)

The app presents three predicted categories and signal visualizations. It is a screening aid, not a diagnostic device, and does not replace professional medical assessment. Screenshots capture the original app locally, not a verified live demo.

SIRAJA empty audio-upload interface with supported formats and preprocessing notes, captured locally.

Actual SIRAJA waveform and spectrogram visualizations from a local app capture; not a diagnostic result.

SIRAJA methodology crop showing CNN-LSTM architecture and 52 MFCC features; performance claims excluded.

(Next project)

Insight-PTN →