How sahana-652/voice-phishing-attack-detector works
This is a voice-phishing detection application. A user selects an audio file in the React interface, submits it to the Flask analysis endpoint, and receives a classification, transcript, and keyword findings for display. The backend combines acoustic features and a loaded classifier with Whisper transcription and keyword checks. A separate training script builds and evaluates a model bundle. The excerpts establish the upload/API and training behaviors, but not every backend call; unsampled connections are omitted. The visible model paths differ between training and serving, so no artifact handoff is asserted.
Main entry points: app.py Analysis API · App.js Detector interface · train_model.py Training pipeline · index.js React entry
Components 14
User interface
- Detector interface
App.js - React entry
index.js
Audio analysis
- Analysis API
app.py - Temporary upload
app.py - Acoustic features
app.py - Speech transcription
app.py - Keyword detection
app.py - Classification
app.py - Model artifact
vishing_model.joblib - Analysis result
app.py
Model training
- Training pipeline
train_model.py - Training labels
labels.csv - Audio sample
1.mp3
Other components
- Audio user
Connections 13
- Audio user → Detector interface: selects audio
App.js - React entry → Detector interface: renders
index.js - Detector interface → Analysis API: posts upload
App.js - Analysis API → Temporary upload: saves file
app.py - Analysis API → Acoustic features: extracts features
app.py - Analysis API → Classification: scales features
app.py - Analysis API → Model artifact: loads model
app.py - Analysis API → Speech transcription: transcribes audio
app.py - Analysis API → Keyword detection: checks transcript
app.py - Analysis API → Analysis result: returns analysis
app.py - Analysis result → Detector interface: displays findings
App.js - Training pipeline → Audio sample: loads audio
- Training pipeline → Training labels: reads labels