Hume AI - The Empathic AI Research Lab | Hume AI

Visit Original

Hume AI is an emotional intelligence lab for voice AI offering open-source models, curated datasets, and evaluation APIs across 50+ languages and 48 emotions, with REST-based human feedback studies, science-backed templates, and models like TTADA, Octave, and EEVI.

  • Hume AI positions itself as an emotional intelligence lab for voice AI, offering open source models, datasets, and evaluation APIs.
  • Its research spans more than 50 languages, 48 emotions, and 600+ voice descriptors.
  • The Human Feedback API provides science-backed survey templates for collecting preference feedback on voice models.
  • The platform supports multiple evaluation dimensions, including listenability, audio quality, and smoothness.
  • Human studies can be run through simple RESTful APIs and use vetted participants from a worldwide pool.
  • Hume claims human preference data can be delivered in hours rather than weeks.
  • The Data offering includes curated speech datasets for tasks such as conversational audio, emotional reproduction, multilingual audio, and voice realism.
  • Datasets also cover domain-specific and task-specific use cases like healthcare, finance, scheduling, support, and onboarding.
  • The models lineup includes TTADA, an open-source LLM TTS system that streams text and audio together to reduce hallucinations and latency.
  • Other models include Octave and EEVI, both closed-source systems with capabilities such as voice design, voice cloning, interruptibility, and expressive instruction following.