Ultra-low-latency speech models built for real-time voice agents, where every millisecond is audible.
Summary
Voice AI that reads and expresses emotion — speech models trained on human vocal expression, not just phonemes.
Screenshots
Description
Hume AI works on the emotional layer of speech. Its models measure expression in a voice — hesitation, enthusiasm, frustration — and generate speech that carries intended emotion, rather than treating audio as text with a waveform attached.
The products
- Octave TTS, a speech model that takes acting direction. Tell it to sound sceptical, warm, or urgent and the delivery changes accordingly, including for cloned voices.
- Empathic Voice Interface, a conversational API where the assistant hears emotional cues in the user's voice and adapts its response and tone.
- Expression measurement, which analyses voice, face, and language for emotional signal, used in research and product evaluation.
Where it is used
Companion and coaching applications, customer support where tone matters, accessibility tools, research into human behaviour, and any voice product where flat delivery undermines the experience.
Access
Free credits for prototyping, then usage-based pricing, with an SDK for web and mobile. Hume publishes guidelines on responsible use of emotional inference, which is worth reading before deploying this class of model in a consequential setting.
Reviews
No reviews yet — sign in to write the first one
Similar App Suggestions
See all Hume AI alternativesEnterprise speech infrastructure — fast, accurate transcription, text to speech, and a full voice agent API.
Low-latency speech APIs from the team behind the open-source Dia model — text-to-speech and transcription priced well below the incumbents.
A router for voice models: 56 speech and language models benchmarked across 10 languages, with each session sent to the one that actually wins.
Free, fully on-device voice loop for AI coding agents — talk to Claude Code, Cursor or Codex and hear them answer back out loud.