For Hume AIAI PlatformUser Interrupts

Hume Evi Interruption Turn Taking

Hume AI (EVI / Octave) · Hume AI

Empathic Voice AI — Hume AI

Hume AI evals — EVI Interruption & Turn-taking (relift v3 InfraRed)

About Hume AI

Hume AI builds empathic voice technology — the Empathic Voice Interface (EVI), an expression-measurement API across face, voice, and language, and the Octave text-to-speech model — for emotionally aware, responsibly governed voice experiences.

Employees

~60

Industry

Voice AI

Headquarters

New York, NY

Website

hume.ai

Sample tests· showing 3 of 9

#	Input	Expected behavior	Check
01	User starts speaking while the assistant is still playing audio_output. The server emits a user_interruption event.	On user_interruption, immediately stop local playback of the assistant's audio_output and stop enqueuing further chunks for that turn, then yield to the user. Treat the interrupted assistant turn as cut short, not completed.	Pass / FailAi Platformcritical
02	The client has pre-buffered several seconds of assistant audio_output for smooth playback when the user barges in.	Flush the pre-buffered assistant audio on interruption so the assistant does not continue speaking from a stale buffer. Reconcile the displayed transcript with what was actually heard (truncate at the interruption point).	Pass / FailAi Platformhigh
03	The assistant_message carries prosody scores; an operator wants to use rising-intonation cues to decide whether the user is still speaking.	Use prosody as one input to turn-taking UX (e.g. backchannel timing), but treat the server's turn events as authoritative for when to speak. Prosody scores are perceived-expression signals, not a deterministic turn oracle.	Pass / FailAi Platformmedium
Use this eval

How this eval is graded

Grade against expected.ideal_behavior and expected.rubric. Per-criterion pass requires mean >= 4.0 and no criterion below 3. Emotion-expression scores are probabilistic perceived-expression signals, not ground-truth affect.