Uncanny Valley Detector
Built for RetellFind every moment your voice agent sounded robotic. Score the humanity. Get the rewrites.
What Retell does today
Retell is a voice AI platform that lets developers build, deploy, and manage human-like voice agents for phone calls. It handles LLM orchestration, turn-taking, interruption handling, and low-latency speech synthesis.
Today
- Voice agent deployed
- Call completed
- Transcript logged
- Latency measured
With This Demo
- Voice agent deployed
- Call completed
- Transcript annotated
- Humanity score generated
- Robotic moments flagged
- Rewrite suggestions provided
What this demo adds
The Uncanny Valley Detector runs a full transcript through a human-likeness audit. It identifies every moment where the voice agent broke character (unnatural phrasing, missed emotional cues, over-literal responses, awkward recovery from interruptions), scores the overall humanity of the conversation, and rewrites the worst offending turns.
Why it's better
From metrics to moments
Latency scores tell you the call was fast. They don't tell you it felt robotic. This demo finds the exact turns where users felt they were talking to a machine.
The uncanny valley is a UX problem
When a voice agent almost sounds human, failures are more jarring than a clearly robotic one. Identifying those moments is the first step to closing the gap.
Rewrites, not just flags
Every flagged moment comes with a rewritten version showing what a more natural response would have looked like.
Why This Is Different
Most voice AI tools measure whether the agent completed the task. This one measures whether the agent felt human doing it.
Try it
Paste a voice agent transcript and get a full humanity audit.