VocEx

Inspiration

When students become freshers in the real world, logging into multiple platforms with passwords and OTPs becomes annoying and unsafe. Passwords can be stolen, shared, or forgotten, and phishing scams are everywhere.

We wanted to solve this everyday problem with something easy to use, secure, and tech-forward. Voice is something everyone has — so why not use it to log in securely? That’s how we came up with VocEx — a simple voice authentication system with real-time spoof detection designed to make login effortless and secure for freshers and everyday users.

What it does

Lets a user log in just by speaking into the mic — no passwords needed.

Checks the voice pattern of the user using speaker embeddings.

Detects whether the audio is real or fake using a machine learning model.

Gives instant feedback: “Access Granted” or “Spoof Detected.”

Works both in the browser and command line — easy to integrate into college platforms, websites, or IoT devices.

How we built it

Frontend: Captured mic input using a simple browser page and Python CLI.

Communication: Used WebSocket to stream audio in real time.

Backend: Built with FastAPI to process voice fast.

Voice Model: ECAPA-TDNN pretrained on the VoxCeleb dataset to create unique voice embeddings.

Spoof Detection: Used a Random Forest model trained on 17 audio features to spot fake voices.

Deployment: Used ngrok to put it online quickly.

We don’t store raw voice — only encrypted embeddings, to keep it secure.

This made the system simple enough for freshers to understand, but powerful enough to solve a real problem.

Challenges we ran into

Making audio stream in real time with low delay.

Getting good spoof detection results with limited data.

Integrating the ML model smoothly with FastAPI.

Handling browser microphone permissions and deployment issues.

Keeping the whole system lightweight so it can run without a GPU — something students and freshers can actually deploy.

Accomplishments we’re proud of

Built a working end-to-end voice authentication system in 48 hours.

Made it beginner-friendly so even new learners can understand and contribute.

Integrated both browser and CLI clients successfully.

Added spoof detection to make the system actually secure.

Delivered a smooth live demo with ngrok — perfect for a hackathon setting.

What we learned

How to stream audio live and process it efficiently.

How to combine speaker recognition with ML spoof detection.

How to design passwordless login flows that feel simple.

The balance between accuracy, speed, and ease of use in a real project.

How to collaborate as a team of students to solve a real-world problem.

What’s next for VocEx

Add multiple user support with a database.

Improve spoof detection with better deepfake models.

Create a simple mobile app so users can log in anywhere.

Add face + voice for extra security.

Make it work offline on edge devices for privacy.

Integrate it into enterprise login systems, job portals, or college attendance apps.

Built With

• css
• fastapi
• html
• javascript
• ngrok
• numpy
• python
• pytorch
• scikit-learn
• sounddevice
• speechbrain