Hume's Audio Decision API lets you ask a recording questions such as "is the caller frustrated?"

Hume AI, which researches how people hear emotion in voices, has launched the Audio Decision API. You send a recording of up to 5 minutes and write questions as sentences, such as "how many speakers are there?", "does anyone laugh?" or "how natural does the agent sound?", each with the answers you allow. It returns a probability for each answer in one call, with no transcript. Access is by waitlist and there's no public price yet.

Why it mattersIt could let a support team flag frustrated calls or test a voice agent without people listening to every recording. But the headline numbers (agreeing with a human panel 70.7% of the time, beating Gemini and Qwen) are Hume's own tests, "10+ models into one call" is a sales line, and long recordings have to be cut into 5-minute pieces. Our check rates it partly fact.

Curated by Sara Simeone as part of our human editorial coverage. The original sources are linked below; curation is distinct from a Hype or Fact? assessment.

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.