Hume measures speech and sound. The Audio Decision API answers questions you write about an audio file and returns probabilities.
Hume AI builds measurement and evaluation for voice. On November 2, 2026 it opened the Audio Decision API. You send one MP3 or WAV, 5 seconds to 5 minutes, and questions written in plain language.
A yes-or-no question returns a probability. A choice returns a probability for each option you list. A scale returns a probability for each level.
A span returns times where an event may occur. The response has no transcript. Docs are at dev.hume.ai.
Billing is $0.47 per million input tokens, covering the audio plus your questions and any context. Answers are not billed. New organizations get a trial credit of 1 million input tokens.
API access is a waitlist at Decision Model Api, which also has a playground for a live clip or an uploaded file. The site also lists Expression APIs, a Human Feedback API, a User Testing API, and Speech Data. Expression Measurement is in limited release and asks teams to contact Hume for access.
Hume's voice docs say the TTS and EVI APIs remain supported until access ends November 13, 2026 at 12:01 a.m. EST, after which account data for those APIs is deleted.
Pricing: Audio Decisions is $0.47 per million input tokens, with no commitment. New organizations get 1 million trial input tokens. A successful decision request is billed for its input tokens. A 4xx rejection, an input-token estimate, and run lookups are not billed. An organization with no payment method, or an unpaid invoice, gets a 402. One that is not permitted to use the API gets a 403 and is told to contact support. Volume pricing, or deployment in your own VPC, on premises, or in country, is Enterprise and requires contacting sales.
Pricing model: Freemium
Common questions about Hume AI
Hume opened the Audio Decision API, which answers questions you write about an audio file and returns probabilities instead of a transcript.
Send one MP3 or WAV, from 5 seconds to 5 minutes, with your questions in the same request. A yes-or-no question comes back as a probability. A choice question returns a probability for each option you listed, such as one speaker, two, or three or more. A scale returns a probability for each level, and a span returns times where something may occur, such…
Other products to explore and follow.
AI media generation platform with 1,000+ top-tier models for fast image, video, and audio generation at scale.
AI singing voice generator — create realistic vocal tracks from text with customizable voices and styles.
One API for 500+ AI models across 80+ providers — pay with credits that work anywhere, with automatic fallback when a provider goes down.
The real-time everything app — posts, news, and conversation, plus the X API and a hosted MCP server for connecting AI agents to real-time data.
Creative platform from ElevenLabs — generate, edit, and localize premium audio and video in minutes with advanced voice, music, SFX, image, and video models.
Lifelike realtime avatar API for conversational AI agents, powered by HeyGen research.
