Hume has introduced an Audio Decision API designed to answer specified questions about speech and sound. The company announcement describes raw audio going in and structured probabilities coming back, rather than requiring a model to generate a long explanation for every judgment.
Developers define the possible answers
A request supplies a question and its allowed answers. The documented forms include yes/no choices, multiple-choice options, rating scales and time spans for events such as laughter or a sigh. Teams can add context and ask several questions about the same recording.
Examples cover speaker counts, background sounds, interruptions and perceived emotional expression. A contact-center workflow could combine these outputs with a transcript and its own business rules. The result remains a model judgment that needs validation for the task and recording conditions.
Perception is different from ground truth
Hume reports 70.7% agreement with a human panel's average judgment in an initial evaluation, compared with 56.5% for an individual-rater baseline. The company explicitly distinguishes relatively objective events, such as a siren, from perceptions such as whether someone sounds frustrated or natural.
Agreement with a panel is therefore not a universal accuracy rate for a person's inner emotional state. Teams should examine the labels, recording conditions and disagreements that matter for their intended use, particularly before letting a score trigger consequential actions.
Availability and a useful first test
The announcement directs developers to a waitlist for API access and a playground for real-time and uploaded-file experiments. It also describes longer-term live and ambient opportunities as real-time capabilities develop, so the broad vision should not be read as a guarantee of every deployment mode today.
A sensible pilot would use representative audio with human-reviewed labels, inspect uncertain answers and test thresholds before connecting the result to an agent. Keep perceived tone separate from the words spoken and from a decision about how a person should be treated.