# Hume AI

> Hume AI is a New York research and technology company that develops emotionally expressive voice models and research infrastructure for voice AI. Its Octave and EVI APIs generate speech and power conversations; its newer focus is training data, expression measurement, simulation, benchmarking, and human feedback for AI labs and enterprises.

- **Founded:** 2021
- **Headquarters:** New York, United States
- **Founders:** Alan Cowen (Founder; former CEO and Chief Scientist; joined Google in January 2026)
- **Products:** Expression Measurement API, Empathic Voice Interface (EVI), Octave, Data Solutions, Kairos
- **Notable:** Announced a $50 million EQT Ventures-led Series B in March 2024., Released the EVI conversation API and Octave expressive speech models., Published customer case studies with GAF and Niantic Spatial.

## Products & services

- **Expression Measurement API** — Originally multimodal expression measurement; current voice-focused offering includes offline Tagger analysis with 600+ dimensions and streaming Prosody measurement.
- **Empathic Voice Interface (EVI)** — Real-time voice conversation API with expressive speech, turn detection, interruption handling, and external LLM integrations. EVI 3 launched July 2025; EVI 4 mini launched October 2025 with an external LLM requirement.
- **Octave** — Speech-language model for expressive text-to-speech, voice design, cloning, and natural-language acting instructions. Octave research preview appeared December 2024; Octave 2 preview added 11 languages in October 2025.
- **Data Solutions** — Licensed voice and multimodal datasets and custom collections, advertised across 50+ languages, 48+ emotions, and 600+ voice descriptors.
- **Kairos** — Agent-to-agent and human-to-agent scenario simulation, evaluation, benchmarking, and regression tracking for third-party voice systems.
- **Human Feedback API** — Managed human studies for live conversation, single-clip ratings, and comparative audio evaluation, returning structured results.
- **TADA** — Open-source speech generation architecture aligning text and acoustic representations; March 2026 release includes 1B and 3B Llama-based models. Assistant scenarios require further fine-tuning.
- **Real World VoiceEQ** — Public multidimensional benchmark and leaderboards spanning speech recognition, speech understanding, text-to-speech, and speech-to-speech.

## Achievements

- Announced a $50 million EQT Ventures-led Series B in March 2024.
- Released the EVI conversation API and Octave expressive speech models.
- Published customer case studies with GAF and Niantic Spatial.
- Open-sourced TADA in March 2026.
- Published the Real World VoiceEQ benchmark paper in July 2026.

## Latest updates

- **2026-01** — Andrew Ettinger became CEO; Hume announced a non-exclusive Google licensing agreement and Alan Cowen’s move to Google.
- **2026-03** — TADA architecture, code, and pretrained models were open-sourced.
- **2026-04** — CEO essay argued that commoditizing voice models shift value toward evaluation, data, and improvement infrastructure.
- **2026-07** — Real World VoiceEQ benchmark paper submitted to arXiv on July 16.
- **2026-09** — September 24 benchmark update evaluated Google Gemini 3.8 TTS preview models with blinded human ratings, and disclosed the Google licensing relationship.

## Links

- Website: https://www.hume.ai/
- LinkedIn: https://www.linkedin.com/company/hume-ai/
- Twitter/X: https://twitter.com/hume_ai
- GitHub: https://github.com/HumeAI
- YouTube: https://www.youtube.com/@hume_ai

---

Profile page: https://yespress.io/hume-ai
Published by YesPress — https://yespress.io
Last updated: 2026-10-03