Is that voice really your customer?
VoiceGuard adds a spoofing and deepfake detection layer to voice authentication, helping banks and call centres stop synthetic or replayed voices.
The problem
Synthetic voices can now fool voice authentication.
AI-generated and replayed voices can trick automatic speaker verification systems. Detection models must keep learning as attack methods change.
accuracy of our MLP classifier on spoofing detection.
random forest accuracy after active learning.
gains in precision, recall and F1 from active learning.
Source: the founding research paper (Journal of Robotics and Control, 2024), using the ASVspoof 2019 database.
How it works
How it works
A clear path from data to decision.
Listen
Audio from the call or app is captured.
Analyse
Classifiers check for signs of spoofing or synthesis.
Learn
Active learning targets the most uncertain samples to improve.
Decide
Genuine voices pass; suspicious ones are challenged.
Product
What makes it different.
Every feature is grounded in the founding research.
Deepfake detection
Identifies synthetic and replayed speech.
Active learning
Improves continuously from the hardest examples.
High accuracy
95.83% with MLP; random forest up to 94%.
Plug-in layer
Adds to existing voice authentication systems.
Who it's for
Who it's for.
Banks
Secure voice-based customer verification.
Call centres
Stop social-engineering attacks by phone.
Telecom providers
Protect voice services from spoofing.
The research
Built on peer-reviewed research.
- Applies active learning to spoofing detection in automatic speaker verification.
- Uses the ASVspoof 2019 database with random forest and MLP classifiers.
- MLP reached 95.83% accuracy; RF improved from 91% to 94% with active learning.
- Precision, recall and F1 improved by 3% to 5% after active learning.
Ahmed, A. S., & Khaleel, A. M. (2024). Enhancing Voice Authentication with a Hybrid Deep Learning and Active Learning Approach for Deepfake Detection. Journal of Robotics and Control, 5(6), 2002–2014.
Team
The team behind it.
Founded in 2024 by a student team from the faculty below.
Faculty of Engineering Technologies (Cybersecurity)
Voice security, machine learning and system integration.
FAQ
Questions we hear often.
Does it replace our voice biometrics?
No. It adds a protective layer that checks whether a voice is genuine.
Can it keep up with new deepfakes?
Active learning lets the model improve as new attack samples appear.
Is this a finished commercial product?
It is a research-based student startup moving from validated prototype toward pilots with real users.
Can we partner or run a pilot?
Yes. We are looking for organisations willing to test the solution in real conditions and share feedback.
Bring VoiceGuard to your organisation.
We are onboarding pilot partners now. Read the research and get in touch.