VOICE & AUDIO · WORTH A LOOK
Hume AI
Voice-AI research company whose new Audio Decision API answers plain-language questions about a recording (is the caller frustrated? how many speakers?) with probabilities.
What Hume AI does
Hume AI builds tools to measure and improve voice AI, based on its research into how people hear emotion and expression. Its developer platform covers voice, expression measurement and, from 2 Nov 2026, the Audio Decision API. You send an MP3 or WAV file (5 seconds to 5 minutes) with up to 25 questions written as sentences, each with the answers you allow: yes/no, multiple choice, a rating scale, or "where in the clip" spans. It returns a probability for each answer, with no transcript or explanation, so your code decides what to do. Uses include scoring support calls, testing voice agents, rating text-to-speech output and searching call libraries. Python and TypeScript SDKs are available.
Good for
- Flagging frustrated callers or interruptions across many support calls
- Rating how natural a voice agent or text-to-speech voice sounds
- Searching a call library for moments such as laughter, sighs or background noise
Try this in 10 minutes
Honest take
No red flags in our notes so far
- Access is by waitlist and there's no public price, so you can't plan costs yet.
- The accuracy and speed figures (70.7% agreement with a human panel, 82% across 10 benchmarks, faster than Gemini and Qwen) are Hume's own and haven't been independently tested.
- It works on files of 5 seconds to 5 minutes. Longer recordings must be cut up, and Hume says live use is still developing.
- Judging emotion from voice is sensitive. Check consent, privacy and employment rules before scoring staff or customers.
Our Hype or Fact? checks
Pricing
No public price. Billed by input tokens (mostly the length of the audio; answers aren't billed). Plans are shown only after you sign in to the Hume Platform. API access is through a waitlist; volume deals via sales.
Prices as listed by the maker when we last checked (2026-11-02); usually in US dollars. Check the website for current prices.
Latest update
- : Launched the Audio Decision API: ask plain-language questions about a recording and get probabilities back. Waitlist access; no public price.