Hume Ai

Hume AI occupies a unique space: AI that measures and responds to emotional expression in voice and face. EVI (Expressive Voice Interface) combines STT, LLM with emotional context, and TTS into a conversational API where the AI adapts based on detected emotional cues. The Expression Measurement API analyzes audio/video/text for facial expressions, tone, prosody, and language. Strong fit for mental wellness apps, customer experience tools, and research that requires affective awareness beyond standard conversational AI. Confidence is docs-derived.

REST API and WebSocket at api.hume.ai. EVI (voice): WebSocket connection for real-time bidirectional voice with emotional context. Expression Measurement: POST /v0/batch/jobs submits audio/video/image/text for async emotion measurement; GET /v0/batch/jobs/{id} polls results. Streaming inference: /v0/stream/models for synchronous small payloads. Emotion outputs: continuous valence/arousal dimensions, discrete emotion labels (anger, joy, sadness, etc.), prosody features.

API key auth: X-Hume-Api-Key header. Keys from Hume AI Platform. HTTPS enforced. WebSocket auth via X-Hume-Api-Key in connection headers. No OAuth2. Rate limiting per plan tier.

Batch job processing: jobs have status (queued, in_progress, completed, failed). Results stored for retrieval. WebSocket reconnection: client responsible for reconnect on disconnect. Expression measurement accuracy varies by input quality (audio/video resolution, lighting). Hume AI uptime at status.hume.ai.

dev.hume.ai/docs provides EVI quickstart, Expression Measurement API guide, model reference, and SDK documentation. Python and TypeScript SDKs available. Getting started: sign up for API access (waitlist or early access), set up API key, first emotion measurement in under 15 minutes. Research foundation: Hume AI was founded on affective computing research.