How sushii-12/voxfusion works
VoxFusion analyzes uploaded or recorded audio for family-emergency voice-cloning scam signals. The web workflow manages family members and reference/sample audio, then runs either an AASIST-only baseline or a combined analysis that gathers speaker, deepfake, transcription, and intent evidence before fusion returns risk and reasons. The README also describes GSM/telephone-codec evaluation. Source excerpts establish the main orchestration and adapter links; some endpoint and storage details are documented but not fully sampled.
Main entry points: services Analysis adapters · main.py FastAPI API · fusion.py Intent and fusion adapters · main.jsx React application
Components 18
Web interface
- React application
frontend/frontend/src/main.jsx
Application backend
- FastAPI API
frontend/backend/app/main.py - Domain records
frontend/backend/app/models.py - SQLite and uploads
frontend/backend/app/db.py - Result schemas
frontend/backend/app/services/schemas.py - Analysis adapters
frontend/backend/app/services
Detection signals
- AASIST adapter
frontend/backend/app/services/aasist.py - ECAPA adapter
frontend/backend/app/services/ecapa.py - Whisper adapter
frontend/backend/app/services/whisper.py - AASIST detection
src/deepfake_detection.py - Speaker verification
src/speaker_verification.py - Audio transcription
src/transcription.py - GSM codec evaluation
scripts
Intent analysis
- Intent and fusion adapters
frontend/backend/app/services/fusion.py - Rule-based intent
src/scam_intent.py - ML intent classifier
src/scam_intent_v2.py
Risk fusion
- Interpretable fusion
src/fusion.py
Other components
- Analyst / user
Connections 17
- Analyst / user → React application: uses
README.md - React application → FastAPI API: calls API
frontend/frontend/src/main.jsx - FastAPI API → Domain records: queries records
frontend/backend/app/main.py - FastAPI API → SQLite and uploads: uses sessions
frontend/backend/app/main.py - FastAPI API → Analysis adapters: runs analysis
frontend/backend/app/main.py - Analysis adapters → AASIST adapter: invokes
frontend/backend/app/main.py - Analysis adapters → ECAPA adapter: invokes
frontend/backend/app/main.py - Analysis adapters → Whisper adapter: invokes
frontend/backend/app/main.py - AASIST adapter → AASIST detection: delegates to
frontend/backend/app/services/aasist.py - ECAPA adapter → Speaker verification: delegates to
frontend/backend/app/services/ecapa.py - Whisper adapter → Audio transcription: delegates to
frontend/backend/app/services/whisper.py - Analysis adapters → Intent and fusion adapters: combines signals
frontend/backend/app/main.py - Intent and fusion adapters → Rule-based intent: analyzes transcript
frontend/backend/app/services/fusion.py - Intent and fusion adapters → ML intent classifier: analyzes transcript
frontend/backend/app/services/fusion.py - Intent and fusion adapters → Interpretable fusion: delegates fusion
frontend/backend/app/services/fusion.py - Analysis adapters → Result schemas: returns typed results
frontend/backend/app/services/aasist.py - GSM codec evaluation → AASIST detection: evaluates codec audio
README.md