What is voice identity for AI agents?
Voice identity tells an AI system who is speaking and whether the voice is live, cloned or replayed — so it can decide whether to act. A practical primer.
Voice identity, explained.
Guides and research on voice identity for AI: voice-clone and replay detection, liveness, speaker verification, voice agent security and on-device Physical AI.
12 articles
How voice cloning attacks AI voice agents — and how to stop it
A few seconds of public audio is enough to clone a voice that passes every knowledge check. How the attack works against AI voice agents, and the defences that hold.
Replay attacks vs. synthetic voices: what’s the difference?
A replayed recording and a cloned voice are different attacks with different artefacts. Why voice security needs both replay detection and synthetic-speech detection.
Voice liveness detection, explained
Liveness detection checks that a real person is speaking now — not a recording, a replay or a clone through a speaker. How it works, passive vs. active, and where it fits.
Speaker verification vs. speaker identification (1:1 vs. 1:N)
Speaker verification checks a claimed identity; speaker identification finds who is speaking among many. How each works, when to use which, and how voiceprints fit in.
Why voice trust must run on-device for robots and Physical AI
Robots and smart devices take voice commands that move things in the real world. Why their voice identity checks must run on the chip — offline, instant, private and secure.
Trusted devices: a second factor for voice commands
For high-risk spoken commands, a phone, watch or earbuds can independently verify the speaker and sign the result. How trusted-device verification works for Physical AI.
How to add voice verification to a Twilio voice agent
Fork the caller’s audio with Twilio Media Streams, score it in real time with the Vosco API, and gate sensitive actions on the trust result. A step-by-step pattern.
Caller verification in Vapi, LiveKit and Pipecat voice agents
Where a voice trust check fits in the three most common voice agent stacks — Vapi tools, LiveKit audio tracks and Pipecat pipelines — plus meetings via Recall.ai and MCP.
The efficiency index: measuring voice security models by size
Accuracy alone hides the cost of running a model. The efficiency index weighs synthesis-detection accuracy against model size — and explains why small models matter.
ReDimNet and the state of speaker recognition research
ReDimNet and ReDimNet2 are open-source speaker-recognition architectures co-authored by Vosco’s team. What they are, why they lead on accuracy versus cost, and why it matters on-device.
Real-time voice trust for contact centers
Score every call for synthetic, replay and speaker risk from the first seconds — before the queue — and keep scoring throughout. How contact centers add voice trust without new friction.
No articles match. Try another search or tag.