Hume AI

Hume AI is a New York-based research lab and technology company developing speech-language models that interpret and generate expressive speech, alongside infrastructure for measuring and evaluating how voice AI systems handle emotional and vocal signals. Its developer platform is delivered through APIs: the Empathic Voice Interface (EVI), a real-time speech-to-speech system that measures users' vocal modulations and generates responses accordingly; Octave, a text-to-speech model the company describes as built on LLM intelligence so that speech reflects the contextual meaning of words; and an Expression Measurement API providing real-time and offline analysis across more than 48 emotion categories, 50+ languages and 600+ output metrics.

More recently the company has positioned itself as the data and evaluation layer for emotionally intelligent voice AI, organized around four product lines: custom data collection (Data Solutions); Kairos, a platform for auto-generating evaluation scenarios and simulating agent-to-agent and human-to-agent conversations to track regressions; the Expression Measurement API; and a Human Feedback API that returns per-sample scores, free-response feedback and aggregated analysis from pre-screened human raters. Hume also publishes public leaderboards, including the Real World Voice EQ benchmark covering speech recognition, speech understanding, text-to-speech and speech-to-speech, and an SLM Judge leaderboard ranking which speech language models most closely track human ratings.

The company's technical work derives from research on how people express and perceive emotion through voice, face and language, including semantic space theory, and it maintains foundational databases built from naturalistic data on more than a million participants. Its stated mission is to ensure AI is built to serve human goals and emotional well-being.

Founding story

Hume AI was founded in March 2021 by Alan Cowen, who serves as CEO and Chief Scientist. Cowen holds a PhD in psychology from UC Berkeley and previously led the Affective Computing team at Google AI, where his academic work established semantic space theory, a computational approach to how voice, face, body and gesture convey emotional meaning. He left Google in 2021 after concluding that voice assistants and large language models lacked sensitivity to human emotion and well-being, aiming to build AI systems optimized for human well-being. The founding team also includes John Beadle (founding investor, CFO and board member), a co-founder and managing partner at Aegis Ventures, and Janet Ho (COO), formerly a managing partner at Aegis Ventures with prior experience at Zynga and Rakuten. The company name references David Hume's argument that emotions drive choice and well-being.

Business model

Hume AI sells access to its models and evaluation infrastructure to developers and enterprises through APIs and a developer platform, with SDKs for React, TypeScript, Python, Swift and .NET and integrations including MCP, Vercel AI SDK, LiveKit, Pipecat, Vapi, Twilio and Agora. Self-serve entry is free to start, with enterprise plans offering custom SLAs, dedicated support and volume pricing; the company also engages in custom data collection, custom integrations and bespoke evaluation work via its research engineering team.

Usage-based API access to speech-to-speech, text-to-speech, expression measurement and human feedback products, with a free tier, paid scaling and enterprise contracts carrying custom SLAs and volume pricing; additional revenue from custom data collection and evaluation engagements.

Traction

By March 2024 the company reported foundational databases with naturalistic data from over a million participants, headcount doubled from 15 to 30, and more than eight academic articles published in top journals over the prior year; a January 2026 profile reported 56 employees and over 40 published articles. In September 2022 it began rolling out a beta to a waitlist of over two thousand companies and research organizations. Published case studies cover Niantic Spatial (AR companions on Snap Spectacles), GAF (training and marketing voiceovers) and Coconote (conversational study tools). Products carry SOC 2 Type II and HIPAA compliance.

Latest developments

Recent releases include the EVI 3 API (July 2025), Octave 2 (October 2025), and Octave 2 in preview together with EVI 4-mini offering expanded language support and lower latency. The company has launched public voice AI leaderboards, the Real World Voice EQ benchmark and an SLM Judge leaderboard, plus the Kairos evaluation and simulation platform and a Human Feedback API, repositioning the company as a data and evaluation layer for emotionally intelligent voice AI.