Expression Measurement
Hume’s Expression Measurement API captures hundreds of dimensions of human expression from audio, video, images, and text. Built on over a decade of research in computational emotion science, these models go beyond basic sentiment to measure subtle expressions like admiration, awe, empathic pain, and dozens more.
Expressions are complex and multifaceted. They should not be treated as direct inferences of emotional experience. To learn more about the science behind expression measurement, visit the About the science page.
Quickstart
Get up and running with Expression Measurement. Each guide walks you through both the Batch API and the Streaming API, from setup to your first predictions.
Use the Python SDK to submit batch jobs and stream predictions in real time.
Use the TypeScript SDK to submit batch jobs and stream predictions in real time.
Use the .NET SDK to submit batch jobs and stream predictions in real time.
Models
Expression Measurement provides a suite of models, each designed for a different modality. You can run multiple models simultaneously on the same input.
48 dimensions of emotional expression from the tone, rhythm, and timbre of speech. Supports audio and video.
48 dimensions of emotional expression from non-linguistic vocalizations like laughs, sighs, and gasps. Supports audio and video.
53 dimensions of emotional expression from the meaning and tone of text. Includes optional sentiment and toxicity analysis.
48 dimensions of emotional expression from facial movements. Optionally outputs FACS 2.0 action units and facial descriptions. Supports images and video.
Each model produces its own set of predictions independently. When you include multiple models in a single job, the response contains separate results for each model.
For example, submitting an audio file with the prosody, vocal burst, and language models enabled returns three distinct sets of predictions, one per model, each with scores for every expression that model measures.
All expression models share a common output format: each expression is assigned a score indicating the degree to which a human rater would identify that expression in the given sample.
See the individual model guides for job configuration options, output details, and the full list of measured expressions.
Batch API vs Streaming API
Expression Measurement is available through two APIs, each designed for different workflows.
Both APIs provide access to the same set of models and return predictions in the same format. Some model job configuration options differ between APIs. See individual model guides for details.
Glossary
Developer tools
Hume provides a suite of developer tools for integrating Expression Measurement.
REST endpoints for submitting jobs, checking status, and retrieving predictions.
WebSocket endpoint for real-time expression measurement.
Official Python and TypeScript SDKs for both batch and streaming workflows.
Open source examples demonstrating batch and streaming integration.
