Blog

Field notes on voice agent testing

Practical writing on how voice calls fail in ways text tests miss, what good scoring looks like, and how to build a testing practice that keeps up with the pace of deployment.

Testing

Why a passing transcript is not a passing call

The transcript says the booking was confirmed. The caller heard three seconds of silence and hung up before the confirmation arrived. Your assertion passed. The call did not.

Nora Halvorsen · July 14, 2026 · 8 min read

From the team

Engineers and researchers who test voice systems for a living, writing about what they find.

Edge Cases

Testing for barge-in: when your agent talks over the caller

Barge-in is one of the most commonly shipped regressions in voice AI. The agent starts speaking before the caller has finished. The caller restarts. The agent interrupts again. Here is how to catch it before it reaches real callers.

Julian Reyes · June 18, 2026 · 7 min read

Scoring

Word error rate is not enough: scoring task success on voice

WER tells you how accurately the transcript matched the audio. It does not tell you whether the call worked. A call can have perfect transcription accuracy and still fail to complete the caller's task.

Nora Halvorsen · May 22, 2026 · 7 min read

Every voice, tested.

Connect one agent and run your first suite of simulated calls in a minute, free.

Run a test suiteTalk to sales