November 2026 Summaries

Hume introduced the Audio Decision API, an audio-native system designed to answer developer-defined questions about speech and sound directly from raw recordings, returning structured answer probabilities and confidence scores. The API supports yes-or-no, multiple-choice, rating-scale, and timestamped span questions covering areas such as emotion, speaker characteristics, language, background noise, non-speech events, conversation dynamics, intent, and communication quality, with questions adjustable without retraining or labeling. Hume positions the tool for real-time voice-agent decisions and large-scale analysis of call archives, where audio judgments can complement transcripts to identify signals such as frustration, uncertainty, interruptions, or customer intent. The model is based on Hume’s emotion-science research and millions of human perceptual judgments; the company reports 70.7% agreement with a human panel’s average assessment in an initial evaluation, compared with a 56.5% individual-rater baseline. Potential applications include evaluating synthetic voices, building more responsive voice agents, and searching or reviewing large audio libraries, while API access is available through a waitlist and a playground offers file-upload and real-time testing.