Introducing the Hume Audio Decision API. Ask anything: How many speakers? Is anyone laughing? Is the caller frustrated? What used to take 10+ models is now one call. Trained on 10M+ human judgments
The Hook
“Alright, so I'm to start over? Yes. Literally once the results come, then like... Hi, I'm Andrew, the CEO of Hume AI, and today we have a really special video for you. One that is not like any that we've ever done.”
Problem Agitation
“AI has been able to see for a long time. However, voice has a listening problem. It reads the transcript, and that fundamentally misses all of the most critical human elements and emotions.”
Product Reveal
“Today, Hume is releasing our Decision API built on the world's first audio-native decision model. It's backed by a decade of research across 48 emotions and 600 different emotional states. It's trained on over 10 million human judgments and is backed by 100 million hours of raw audio.”
Feature Teaser
“So 50 different languages, you're now able to ask questions—one, ten, twenty at once—and get back answers quickly. Normally you'd have to stitch together ten different models... you can just stitch them all together in one call and get an answer to all of your audio questions basically under a second.”
Feature Teaser
“Hey, here's a quick demo of how Hume's new audio understanding model works... It does support many different accents now... It says Korean right there, and also Urdu... Something cool here too is like you can just dynamically add more...”
Call to Action
“(No spoken dialogue — ambient modern synthesizer outro music)”
Synthesizable in Remotion · key components of the audio-decision-api-launch template.
Low-opacity monochrome dot-matrix world map background with pulsing regional luminance.
Render an SVG/Canvas grid of squares; evaluate Perlin noise offset by current frame to modulate individual square opacity.
Dark translucent glass panel displaying query title, percentage meter, and classification tag.
Card component using Remotion spring() for entrance scale/opacity; animate child progress bars with interpolate(frame, [0, 20], [0, targetVal]).
Horizontal scrolling multi-track audio waveform indicating active speech, latency, and turn boundaries.
Map audio amplitude data array to vertical SVG lines; translate X-position linearly using frame * speedFactor.
Purpose. Establish authenticity and team credibility before product claims.
Execution. Observational footage of team blooper in glass conference room.
Purpose. Identify core limitation of transcript-based voice AI.
Execution. Schematic graphic contrasting text transcript with vocal nuances.
Purpose. Unveil Hume Decision API with empirical scale metrics.
Execution. Animated data tickers over dark dot-matrix background.
Purpose. Prove speed, accent handling, and multilingual support live.
Execution. Direct screen captures alongside live team testing.
Purpose. Direct developers to documentation and demos.
Execution. Minimalist typography with subtle glitch transition to logo.