Hume AI

Hume AI works on voice and emotion research. It offers models that judge audio, and tools that measure how voice AI performs in real conversations. Its products are aimed at AI labs and companies that build voice products.

The company's audio-native decision model answers questions about a recording, such as the emotion being shown or the language spoken. It returns structured, probabilistic judgments without needing a transcript. Teams can use it to evaluate voice agents on tone, timing, interruptions and context, and to guide improvements. The site also offers a demo and a talk-to-us route for teams.

• Audio questions answered as structured results
• Emotion and tone judgments
• Evaluation of voice agents
• Timing and interruption checks
• API access for teams

AI labs, voice product teams and companies that run customer calls and need to test how a voice system sounds in practice. Researchers in emotion and speech may find the benchmarks useful.

Check the pricing page for current access options. Judgments come with probabilities, so treat them as guidance rather than fixed facts. Human listeners are used as a reference in the company's research, which is one reason its results are presented as probabilities.