Today we just shipped Arcana, and we think it’s the most realistic spoken language model yet! Infinite custom voices from just fictional names or simple descriptions. Laughter and sighs plus dynamic emotions. Multi-lingual out of the box.

Universal-3 Pro by AssemblyAI — Speech-to-text that finally understands context

Speech-to-text that finally understands context

Just launched: Arcana – the most realistic TTS model ever built 🧠🔊 We've spent the last year pushing the boundaries of neural speech synthesis, and Arcana is the result. It’s not just lifelike—it’s emotionally nuanced, responsive to context, and capable of sounding eerily human. From whispered asides to emphatic shouts, Arcana gets it right. Whether you’re building character-driven games, creating synthetic voiceovers, or designing AI companions, Arcana lets you bring voices to life with unprecedented realism. We’re so excited to finally share this with the world. Would love your thoughts, questions, or wild use cases—we’re here all day! 🎧 Try a demo, give it a spin, and let us know what you think.

Universal-3 Pro by AssemblyAI — Speech-to-text that finally understands context

Speech-to-text that finally understands context

Arcana: AI Voices with Vibes

The most realistic text-to-speech model ever

The most realistic text-to-speech model ever