Hume launches Audio Decision API that answers questions about raw audio

Hume AI has introduced the Audio Decision API, which it describes as the first audio-native model that answers custom questions about speech and sound. Developers send raw audio plus a question they define, such as whether a voice sounds frustrated, whether it is AI generated, or whether a siren is in the background, and the API returns a structured answer with a confidence score. Questions can be yes or no, multiple choice, rating scales, or spans that mark when an event like a laugh or sigh happens.

Hume says the system is fast enough to ask dozens of questions every few seconds on a live call, and to make an archive of 40,000 recordings searchable within minutes, with no labeling, retraining or transcript pipeline. Adding a new question is a line of JSON. The model builds on Hume's emotion science research and millions of human judgments about how people perceive speech and sound.

In an initial evaluation, Hume says the model reached 70.7 percent agreement with a human panel's average judgment, versus 56.5 percent for an individual rater, and it claims better accuracy, speed and cost than existing audio methods based on its own benchmarks. Access is through a waitlist, with a playground for trying it.

Why it matters

It gives developers a way to query emotional and acoustic cues directly from audio without building separate classifiers or transcription pipelines for every new question.

Source: Hume AI