Full-Code · Voice & Conversationalactive

Hume EVI

also known as Empathic Voice Interface, Hume Speech-to-Speech

Type: full-code · Vendor: Hume AI · Language: API · License: proprietary · Status: active · Status in practice: emerging

Links: homepage docs

Hosted speech-to-speech voice API from Hume AI that pairs an emotionally aware response model with a configurable supplemental LLM, measuring vocal prosody and adapting tone in real time.

Description. EVI is Hume AI's real-time speech-to-speech interface. It streams measurements of the tune, rhythm and timbre of the user's voice, reacts with matching prosody, and remains interruptible at all times. EVI is configured through a Config object (system prompt, voice, supplemental LLM, tools) and can be supplemented with partner LLMs from Anthropic, OpenAI, Google or Fireworks. Tool use and built-in tools are first-class but parallel function calls are not yet supported. Chat Groups link sessions so a conversation can resume across disconnects.

Agent loop shape. Hosted speech-to-speech loop over a WebSocket. Caller audio streams in; EVI emits prosody measurements, decides the response (optionally via a supplemental LLM), streams generated speech back, and is interruptible by design. Tool calls go out to the developer's backend except for Hume's built-in tools which it invokes itself. A Chat Group id can be passed to resume across reconnects.

Primary use cases

emotionally aware voice agents across consumer and support apps
multilingual speech-to-speech experiences on EVI 4 / 4-mini
function-calling voice agents that hit external APIs
long-running conversations resumed via Chat Groups

flowchart TD user[User audio] --> ws[WebSocket] ws --> evi[EVI speech-to-speech] evi --> prosody[Prosody measurements] evi --> resp[Response generation] resp --> llm{Supplemental LLM?} llm -->|yes| partner[Anthropic / OpenAI / Google / Fireworks] llm -->|no| native[Native EVI response] partner --> tool{Tool call?} native --> tool tool -->|developer fn| backend[Developer function] tool -->|built-in| builtin[Hume built-in tool] tool -->|none| tts[Streaming TTS reply] backend --> resp builtin --> resp tts --> user interrupt[User interjects] -.barge-in.-> tts group[Chat Group id] -.resume.-> evi

Key concepts

EVI speech-to-speech (docs) — Streaming voice loop that measures vocal prosody and matches tone.
Configuration (docs) — EVI Config defines system behavior, voice and supplemental LLM.
Supplemental LLM (docs) — Optional partner LLMs from Anthropic / OpenAI / Google / Fireworks plug into the loop.
Tool use → tool-use (docs) — Developer-defined functions and Hume built-in tools; no parallel calls.
Multilingual EVI 4-mini → multilingual-voice-agent (docs) — Languages: English, Japanese, Korean, Spanish, French, Portuguese, Italian, German, Russian, Hindi, Arabic.
Chat Groups → agent-resumption (docs) — Link related chats so a conversation persists across disconnects.