Accelerating Trustworthy Voice AI Adoption.
Public benchmarks for every layer of voice AI. MIVAS evaluates speech-to-speech models, MISTT evaluates speech-to-text, and text-to-speech is next.
01 / Mission
Our mission is to accelerate the adoption of safe, trustworthy, and reliable conversational intelligence systems.
Bluejay Labs exists so voice systems can be used in the world, not only capable in a demo. Safe, trustworthy, and reliable enough to adopt.
02 / Method
We accomplish this by building RL environments and industry benchmarks to steer models towards desirable outcomes for humanity.
Worlds, tools, and tasks with verifiers that train the behavior. Public scores then show whether those systems are getting better.
03 / Datasets
Our datasets are designed for speech-to-speech voice AI models and conversational intelligence systems.
Production-shaped voice turns: overlap, tools, and real task pressure. Not generic text scraped from the web.
04 / Benchmarks
Benchmarks
MIVAS
Speech-to-speechSpeech-to-speech models run stateful voice-agent tasks in healthcare, legal, and customer support, scored by deterministic verifiers.
- Models
- 13
- Tasks
- 216
- Industries
- 3
MISTT
Speech-to-textSpeech-to-text models on clean, phone-line, noisy, and crosstalk audio, with streaming latency and confidence intervals.
- Models
- 81
- Clean clips
- 2,999
- Vendors
- 25
TTS
Text-to-speechA text-to-speech benchmark for voice agents is in development, completing the cascaded stack alongside MISTT.
Coming soon