Dessa comes from Block’s research efforts and is presented as a research preview of a conversation model built exclusively on speech. Rather than converting audio to text and back again, it works directly with spoken human conversation, aiming to make voice interaction feel closer to talking with a person.
The core idea is full-duplex speech modeling: the system listens and responds in real time while retaining paralinguistic cues such as emotion, nuance, pauses and natural cadence. Because it was trained from scratch on human conversation without text or pretrained models, the team positions it as a new category of model with behaviors researchers can study, including how efficiently language and facts are learned from audio alone.
It suits AI researchers, speech and audio engineers, and product teams investigating voice-first interfaces, as well as curious early adopters who want to test a live spoken conversation rather than a scripted demo. Those needing production-ready voice agents should treat it as exploratory.
Access is currently handled through a waitlist, and a download option is offered on the page. Interested users should join the waitlist and follow the published research materials before planning any integration, since capabilities and availability are still evolving.
In short, Dessa is a focused experiment in speech-only conversational AI, best approached as a research preview rather than a finished product.
Comments & Ratings
0No comments yet. Be the first to review.