System Architecture

Processing Pipeline

Audio Input
    |
    v
Audio Preprocessing
    |
    v
Feature Extraction
    |
    +--> MFCC
    +--> Mel Spectrogram
    +--> Spectral Features
    +--> Pitch
    +--> Zero Crossing Rate
    +--> Chroma
    +--> RMS Energy
    |
    v
Feature Aggregation
    |
    v
Feature Vector
    |
    v
Machine Learning Classifier
    |
    v
Prediction
    |
    v
REAL / FAKE
+ Confidence Score

---

Processing Modes

S.no.	Modes	Behaviour
1.	Uploaded Audio	The user uploads an audio file through the frontend.
2.	Browser Microphone Input	The browser captures microphone audio and converts it to WAV before sending it to the backend.
3.	Backend Inference	The Flask API extracts features using the same extraction process used during training and passes the resulting vector to the trained model.

---

Design Principle

The feature extraction logic must remain consistent across:

• Model training
• Local microphone detection
• Backend prediction

Training and inference must use the same feature order, dimensions, and preprocessing assumptions.