Clinician and caller simulator for voice
Plays callers with accents, noise and interruptions.
Checked against clinicians
Realism and WER-matching against real calls.
Where a person decides
Same; no real-patient contact.
Works on
Voice agents, scribes (ASR).
In development. Early access is open. Reviewed .
The problem it solves
Voice agents meet accents, background noise and interruptions. Lab recordings with clear speech don't expose these failures.
How it works
Create scenarios
Scenarios cover booking, refills and symptom intake.
Add real-world conditions
Accents, noise, pauses and interruptions are layered in.
Repeat and score
Each scenario runs several times. It passes only if every run passes.
An example, illustrative
Clean call: books the visit.
Noisy call with an interruption: books the wrong day. Fails 2 of 5 repeats.
A phone or API endpoint for the voice agent.
Synthetic callers can't cover every voice. We check realism against real calls.
What it helps measure
Quick answers
What does the caller simulator do?
Simulate callers with accents, noise, interruptions.
How is the caller simulator checked against clinicians?
Realism and WER-matching against real calls.
Can it do Indian languages?
That's the goal, with native-speaking clinicians reviewing. See the language evaluation service.
Why repeat each call?
Voice agents vary between runs, so one pass doesn't prove reliability.
What can the caller simulator not do?
Same; no real-patient contact. No agent makes a clinical decision.