Expression Measurement API

Identify emotion in voice, offline or in real time

The Tagger API returns 600+ expression and emotion dimensions from any audio, at scale. The Prosody API delivers real-time emotional signals during live conversations. Both built on the same expression science.

Two ways to use it

Offline analysis or real-time signals, same model, different delivery

Tagger, Offline

600+ expression dimensions. Batch processing. Deep analysis.

Send audio, get back a rich emotional profile across 600+ dimensions spanning emotions, speaking styles, vocal qualities, and expression intensity. Built for training-data annotation, model-output analysis, and large-scale evaluation runs.

Prosody API, Real time

Emotion signals during live conversations. Sub-second latency.

Stream audio and receive real-time emotion identification as a call unfolds, capturing frustration, distress, warmth, and engagement. Built for voice-native AI providers who need to respond to how callers actually feel, as it happens.

What gets tagged

600+ dimensions across every layer of human expression

Discrete emotions

AfraidAngryAmusedJoyfulDisgustedExcitedDistressedSadSurprised

Speaking styles

WhisperCalmExcitedNervousAssertive

Vocal qualities

PaceWarmthEnergyClarityHesitation

Expression intensity

Per-dimension confidence scores on every tag.

How teams use it

From training data to production monitoring

Training data annotation

Tag large audio datasets with emotional labels automatically, at the scale model training actually requires.

Model output evaluation

Understand the emotional profile of your model's generations. Know whether outputs are calibrated to context before a training run ships.

Real-time routing & response

Detect caller frustration or distress mid-call and route or adapt accordingly, without waiting for the conversation to end.

Production monitoring

Track whether deployed voice AI is responding appropriately to caller emotion over time. Catch drift before it becomes a complaint.

Built on expression science

The most comprehensive emotion taxonomy in voice AI, built on decades of research

600+ dimensions

The most complete expression taxonomy available, covering discrete emotions, speaking styles, and vocal qualities across every register of human speech.

50+ languages

Trained and validated across linguistic and cultural variation, because emotional expression isn't universal, and the model knows it.

Proven in production

Used internally to evaluate Hume's own voice AI across training, alignment, and release, the same model that powers our own evals.

Start tagging emotion in your voice data

Contact our team for API access, or learn how the Prosody API integrates into your live conversation stack.

Stay in the loop

Get the latest on empathic AI research, product updates, and company news.

Join the community

Connect with other developers, share projects, and get help from the team.

Join our Discord