Hume EVI
Empathic Voice Interface: conversational AI that hears and responds to emotion, sub-300ms.
About
Hume's Empathic Voice Interface (EVI) is a speech-to-speech API for building voice agents that detect and respond to emotional expression. EVI 3 runs with sub-300ms latency, supports interruptible conversation, custom voice personas, and multilingual speech, while the Expression Measurement API reads 25+ emotions from voice, video, and text. It ships alongside Octave, Hume's LLM-native TTS. Google DeepMind licensed Hume's emotional capabilities in early 2026, a strong signal of the tech's standing.
This is an API-first developer product, not a consumer app — you write code to use it. Pricing is usage-based (EVI minutes billed per tier, roughly $0.04–$0.07/min overage), so high-volume deployments need cost guards. Non-English support is narrower than the big TTS vendors, and raw voice fidelity still trails ElevenLabs for pure narration use cases.
What it can do
- Speech-to-speech with emotion understanding
- Sub-300ms conversational latency
- Custom voice personas from text or audio
- Expression Measurement API (25+ emotions)
- Octave LLM-native text-to-speech
- Interruptible, natural turn-taking
Setup
Create an account at hume.ai and generate an API key. Install the Hume SDK (Python/TypeScript) or hit the REST/WebSocket endpoints directly. Free tier: 5 EVI minutes/month; paid tiers from $3/mo Starter.
Prerequisites
Developer setup (API key, SDK); a language model of your choice (bring your own key or use Hume-managed).
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 51; Trending score 81/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.