Hume AI

Hume AI is the Emotional Intelligence Lab for Voice AI, providing open-source models, datasets, and evaluation APIs to embed emotional intelligence into voice models. Our research spans 50+ languages, 48+ emotions, and 600+ voice descriptors, and we offer human-feedback tooling for evaluating and improving models.

Features

• Open-source LLM TTS system TADA that streams text and audio, reducing hallucinations and latency.
• Closed-source LLM TTS system Octave with voice design, modulation, voice cloning, and voice conversion.
• Closed-source LLM Speech-to-Speech system EVI with interruptibility, back channeling, expressive instruction following, and external LLM compatibility.
• Human Feedback API and Study Runner to run scientifically grounded preference evaluations; RESTful API to create and manage evaluation studies.
• Data Library with curated speech datasets spanning 50+ languages and diverse domains, including conversational audio and domain-specific data.
• 48 core emotions and 600+ voice descriptors across multilingual data.

Who Is It For?

• Researchers and developers building empathic voice AI.
• ML engineers and data scientists evaluating and improving voice models with human-in-the-loop feedback.
• Organizations needing multilingual data and evaluation tooling for product and research teams.

Use Cases

• Build and deploy empathic voice agents with validated emotional intelligence.
• Train on curated multilingual datasets across healthcare, finance, and more domains.
• Run rapid human evaluations via the Human Feedback API and Study Runner.
• Benchmark models using evaluation APIs and RESTful evaluation tooling.
• Fine-tune or contrast performance with domain-specific datasets.

---

Get started with empathic voice AI today.