Jack Dorsey's Block released Dessa, a speech-to-speech model trained only on recordings of human conversation, with no text and no pretrained models. It is full-duplex, meaning it can listen and talk at once and handle interruptions and short acknowledgements the way people do. The company says it runs locally on a MacBook Pro with an M4 Max chip or newer. It is not yet publicly available: Block is giving early access to a small group of AI safety researchers and has opened a waitlist.
What it means: Voice systems that talk naturally, with pauses and interruptions, change how businesses can serve customers on the phone. For Greek call centres, hotels or shops that take bookings and orders, a model like this could one day handle simple calls more naturally than today's touch-tone menus. Dessa was trained on conversations, though, and it is not known how well it covers Greek, so it is not immediately usable here. The bigger issue is security: the more natural a machine voice sounds, the easier scams such as 'I am calling from your bank' or 'this is your director' become. Businesses should already have rules so that no payment or change of bank details is approved on a phone instruction alone, with confirmation through a second channel. If a company does use such a system to talk to customers, the EU AI Act requires telling people that they are speaking with a machine. The fact that the model runs on a laptop also shows that voice AI does not always need the cloud, which helps applications that must keep recordings on local machines for privacy or regulatory reasons.
Related service: Custom software
Related stories
- OpenAI introduces GPT-6 Cyber for aligned cybersecurity research
- NVIDIA Cosmos offers open tools and checkpoints for interactive world simulation
- Mistral Studio expands with managed agents, workflows and enterprise apps
- Google launches Gemini Omni 1.2 Flash with 15-second video and 30% lower price
- Dropbox plans to buy Tigris and launches DBX Agent Cloud for AI storage