Block research preview of a full-duplex speech-to-speech model trained only on human conversation audio, without text.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Dessa is a research-preview speech-to-speech model from a small AI research lab at Block, founded by former members of the Toronto deep learning startup Dessa. According to Block, it was trained from scratch on human conversational speech only, with no text and no pretrained models, and learned to hold real-time conversations by listening.
The model is full-duplex: it listens and speaks at the same time, decides when to interrupt or backchannel, and produces emotion, laughter, sighs, breathing and accents directly in audio rather than through a text intermediate. Block says inference runs faster than real time and that the model can run locally on a MacBook Pro M4 Max or better. It reports training on a single NVIDIA GB200 NVL72 and using far fewer words than text LLMs need to learn language.
Access is limited. Block is offering early access to a small group of AI safety researchers to look for unanticipated risks; there is no public product, API or pricing described. Demo recordings are published on the website, including a conversation with a third-party model, GPT-Live-1. Claims such as being the first or most authentic such model are the developer's own.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Real-time speech-to-speech voice model that learned from human conversations without text, picking up accents, breathing and humor. Access is limited to a small group of AI safety researchers.
Closest neighbours by meaning, not just by category.
Model · Thinking Machines
Thinking Machines' preview real-time model that processes audio, video and text continuously while responding.
Model
Full-duplex real-time voice, video and screen assistant from MOSS (模思智能) with background task execution.
Model · Alibaba
Alibaba research model for real-time conversational avatars using one transformer over text, audio and video.
Model · Microsoft
Microsoft AI's text-to-speech voice model with expressive output across multiple languages.
Model · Alibaba
Alibaba's real-time audio-visual interactive character video model that renders speech and free-form actions.
Model · xAI
xAI's voice model for real-time spoken conversations and voice agents, available through Grok apps and the xAI API.
