Crédit Agricole × Sopra Steria hackathon
Fight the Fraud
Checking by voice that the client is talking to their real advisor.
- Python
- SpeechBrain
- React
- FastAPI
- Period
- October 2025 · Lyon
- Team
- Team project
- Code
- View the code
The project
A hackathon about fighting bank fraud by processing client data with AI.
Our project: voice recognition during calls with an advisor, so that clients know they are talking to the right person. The voice is compared to a reference sample with SpeechBrain, a second model detects synthetic voices, and a React interface shows the result.
Try it.
The audio pipeline
Step 1
The call
The advisor’s audio and the client’s, cut into 10-second segments.
Step 2SpeechBrain
Voice verification
The advisor’s voice is compared with a reference sample (ECAPA-TDNN model).
Step 3Transformers
Synthetic voice
A second model detects AI-generated voices.
Step 4n8n
Conversation analysis
The content of the exchange is analysed to spot suspicious requests.
Step 5React
Result
The interface shows the match score and the alerts.