User Testing API
Put your voice AI to the test with real people.
Run live tests of your voice model or agent through our API, with qualified participants matched to your testing brief. Get recordings, transcripts, and feedback to guide your next iteration in hours, not weeks.
Scenario: cancel a plan, then change your mind
- Recordings
- Transcripts
- Feedback
- 96
Users in 50+ languages
English, Spanish, Mandarin, Hindi, Arabic, Portuguese, French, German, Japanese, Korean, Indonesian, Bengali, Turkish, Vietnamese, Italian, Polish, Dutch, Thai, Tagalog, Swahili, Urdu, Ukrainian, Malay, Romanian, Greek, Czech, Hebrew, Swedish
Reach the right participants. Leave the logistics to us.
Set participant criteria, scenarios, and feedback questions for scripted, guided, or open-ended conversations. Bring your own scripts or generate them with Hume, with recruitment, screening, consent, payment, and session quality handled for you.
From prototype to production. Get results in hours.
Connect your endpoint to evaluate responses, timing, turn-taking, and fallbacks. Get the feedback you need to keep development moving.
wss://agent.yourco.com/v2
- Responses
- 0.82
- Timing
- 0.64
- Turn-taking
- 0.58
- Fallbacks
- 0.71
- Brief launched
- First sessions complete
- Results ready
Create data for your next release.
Each run evaluates how your model handles real people and creates a corpus of realistic conversation data you own and can train on. Get both from consented test sessions, without using production PII.
- Consented
- No production PII
- Yours to train on
Actually, wait. Can I keep the plan if I drop the extra line?
Your cancellation is confirmed.
- Missed interruption
- 01:52
- Late turn-taking
- 02:10
- Participant rating
- 2 / 5
- session_0142.wav
- transcript.json
- labels.json
Used for
Find the gaps that real conversations reveal.
Release testing
Find where your system struggles with interruptions, hesitation, confusion, and unexpected requests before customers encounter those failures.
Market expansion
Evaluate how your voice AI performs across languages, accents, and audiences before entering a new market.
Training data
Build a corpus of realistic, consented conversations for evaluation and training from the same sessions you use to test performance.
Built for voice, from research to infrastructure.
Voice-native infrastructure
- From audio processing to rater recruitment, every layer of our infrastructure is purpose-built for voice and emotion.
Trusted by leading AI research labs
- Hume’s research and infrastructure support teams building and evaluating voice AI.
1 million+ human ratings
- Behind Real-World VoiceEQ, one of the largest human evaluations of voice AI.
Start with your endpoint and a testing brief.
Define who should participate, what they should attempt, and what you want to learn. Launch testing through the API, with Hume handling recruitment and session operations.