Voxmind builds embeddable voice biometrics detection models targeting sub-500MB of runtime memory to fit inside hardware rather than beside it in data centers.
Published
Signal category
Products & Services
Quote
“That is why we build Voxmind to be embeddable, targeting sub-500MB of runtime memory so it fits inside the hardware already sitting on desks rather than beside it in someone else's data centre.”
— Jaikiran Keerthi|Voxmind team
Company
Voxmind
"Your Voice, Your Identity - Instantly Recognised"
- Industry
- Technology, Information and Internet
- Location
- London, GB
- Company size
- 7 employees
Voice is the last channel enterprises still trust by ear. Voxmind makes it verifiable. We build real-time voice deepfake detection and authentication grounded in the physics of human speech. Instead of pattern-matching against known fakes, our engine analyses the phoneme-frequency relationships only a real human vocal tract can produce. That means it holds up against new synthesis models without retraining, works across languages by design, and runs live in the call: cloud, API or fully on-device. And because we detect authenticity rather than identify speakers, there is no enrolment and no stored voiceprint. Nothing biometric to breach. Banks, telcos and contact centres use Voxmind to stop voice fraud before an agent, a customer or a transaction is exposed. Benchmarked on ASVspoof 2021: 0.78% EER. Verification in under three seconds.
Founded 2024