Emotion Recognition API
The AI USM emotion recognition API exposes the same multimodal analysis that runs behind the assistants: send text, audio or visual input and receive an estimate of emotional context with per-channel confidence, rather than a single unqualified label.
Input modalities
The API accepts the channels the platform itself uses. Requests may carry one modality or several; sending more than one is what makes fusion possible.
- Text emotion analysis — conversational or document text
- Audio emotion analysis — speech, analysed acoustically
- Visual analysis — frames or short sequences from a camera
- Multimodal requests — channels combined into one estimate
What a response contains
A response describes the estimated emotional context together with the confidence attached to each contributing channel, so an integrating system can decide how much weight to give it. Where a channel is missing or degraded, that is stated instead of silently absorbed.
Intended use and limitations
Intended for emotion-aware interfaces, support and wellbeing products, research, and human-reviewed workflows. Not intended for surveillance, automated screening of individuals, employment decisions, or any use that treats an estimate as a fact about a person. The API does not produce medical diagnoses.
Getting started
Technical detail, request format and integration notes are in the documentation; commercial terms are on the API pricing page.