Hume AI

What Hume AI does

Hume AI builds the data and evaluation layer for emotionally intelligent voice AI. Built from a decade of voice and emotion research, Hume provides the research infrastructure behind expressive, trustworthy voice AI, including scientifically grounded datasets, speech models, evaluation frameworks, and human preference pipelines. They help teams measure, improve, and align voice AI with how people actually experience conversations, whether building foundation models, fine-tuning voice agents, or evaluating production systems.

Products

• Data SolutionsCustom data collection built around specific scenarios, use cases, and evaluation requirements for models.
• Kairos PlatformAllows voice AI teams to build evaluation suites from real-world use cases, simulate agent-to-agent and human-to-agent conversations, and track regressions over time.
• Expression Measurement APIMeasures expression in voice in real time, offering emotion tagging and offline analysis across 48+ emotions and 50+ languages with 600+ output metrics.
• Human Feedback APIProvides pre-screened human raters for per-sample scores, free-response feedback, and aggregated analysis in hours.
• RW-Voice-EQ BenchA benchmark with four leaderboards (Speech Recognition, Speech Understanding, Text-to-Speech, Speech-to-Speech) ranking voice AI models based on human judgment across recognition, understanding, expression, and conversation.
• SLM JudgeA leaderboard for judges (SLMs) to track which automated evaluators most closely align with human ratings.
• Conversational AIDrop-in conversational AI for applications, enabling AI agents to engage in natural, emotionally intelligent conversations, understand context, adapt to user sentiment, and respond with empathy.
• Expressive Text-to-SpeechGenerates speech with natural emotion, allowing control over tone, pacing, and emotional delivery to create genuinely human-feeling voice experiences.
• Voice ConversionTransforms any voice into another while preserving emotional depth, intonation, and expressiveness.

Key capabilities

• Custom data collection
• Evaluation suites from real-world use cases
• Agent-to-agent and human-to-agent conversation simulation
• Real-time emotion tagging and offline analysis
• 48+ emotions and 50+ languages supported
• 600+ output metrics
• Pre-screened human raters
• Per-sample scores and free-response feedback
• Aggregated analysis
• Voice AI leaderboards (RW-Voice-EQ Bench)
• SLM Judge leaderboard
• Single API call for study creation to results
• Sophisticated participant screening, fraud detection, and quality monitoring
• Hours, not days, for ratings feedback
• Custom integrations, evaluations, and question types
• SDKs for TypeScript, React, Python, Swift, .NET
• Integrations with Agora, LiveKit, Twilio, VAPI, Vercel AI SDK
• Voice cloning
• Commercial license for voice conversion
• Team collaboration and unlimited team seats
• SOC 2 Type II, GDPR, HIPAA compliance

Use cases

• Building foundation models
• Fine-tuning voice agents
• Evaluating production systems
• Measuring and improving voice AI
• Bringing digital characters to life with emotionally rich voices
• Transforming AI phone support
• Professional training
• Professional audio for creatives
• Giving enterprise AI agents a voice
• Creating immersive avatar experiences
• Creating interactive and spatially aware AI companions
• Detecting voice fraud in real-time
• Creating podcasts at scale
• Conversational learning
• Preventing harmful deepfakes
• Generating expressive text-to-speech
• Building AI agents that engage in natural, emotionally intelligent conversations
• Integrating voice and chat experiences that understand context and adapt to user sentiment
• Transforming voices while preserving emotion

AI approach

Hume AI builds the data and evaluation layer for emotionally intelligent voice AI. They develop research infrastructure including scientifically grounded datasets, speech models, evaluation frameworks, and human preference pipelines. They use data-driven methods, computational techniques, and large-scale human studies to understand how people express and perceive emotion across voice, face, and language, which powers their AI infrastructure.

Tech named: speech models, evaluation frameworks, human preference pipelines, multilingual datasets, public benchmarks, feedback systems

Industries served

• AI Companions
• Phone Support
• Professional Training
• Audio Production
• Enterprise AI
• Gaming
• Fraud Detection
• Podcasting
• Education

What it says sets it apart

• Built from a decade of voice and emotion research
• Scientifically grounded datasets, speech models, evaluation frameworks, and human preference pipelines
• Focus on emotional intelligence, expressivity, naturalness, flow, reliability, safety, and trustworthiness in voice AI
• Metrics measure accuracy, while human judgment and tools collect expression, emotion, and alignment
• Single API call for study creation to results, no operational burden
• Sophisticated participant screening, fraud detection, and quality monitoring built-in
• Agent-to-agent and human-to-agent conversation simulation for real-world scenarios
• Ratings returned in hours, not days, to close human evaluation loops at model-development pace
• Proven on real models, used internally for evaluation across simulation, expression measurement, and human studies
• Custom integrations, evaluations, and question types with research engineering team support
• Full spectrum of emotion in voice AI, not limited to basic emotions
• Data-driven methods, computational techniques, and large-scale human studies to understand emotion perception

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.