Hume AI

Hume AI builds voice models that aim to sound emotionally expressive rather than flat, and it provides tools to measure how well a voice AI actually comes across to real listeners. Its two headline products are Octave, a text-to-speech system that generates a voice and personality from a written prompt or a short recording, and EVI (Empathic Voice Interface), a real-time speech-to-speech model, now in its third version, that picks up on a speaker's vocal cues and responds with matching expressiveness. Both are built on language-model intelligence rather than older signal-processing approaches, which the company says lets them understand what words mean in context instead of just how to pronounce them. Beyond voice generation, Hume also sells evaluation infrastructure: an Expression Measurement API that scores emotion across dozens of categories and many languages, a Human Feedback API that returns ratings from screened human panels, and a scenario-simulation platform called Kairos for building test suites. This evaluation layer is aimed at teams building conversational or voice products who need a way to judge quality beyond standard accuracy metrics. The company publishes developer documentation and public leaderboards comparing voice models and evaluator quality. Pricing for the evaluation services is not published and appears to require direct contact, which limits how easily smaller teams can estimate costs upfront.