"Block" reveals AI model that speaks, laughs, and interacts like humans

American company 'Block' has unveiled a new voice AI model named 'Disa', capable of simulating human conversations including laughter, pauses, and tone changes, after being trained from scratch on recordings of human conversations.

American company 'Block', led by Jack Dorsey, has unveiled a new voice AI model named 'Disa', capable of conducting conversations that mimic human speech, including laughter, pauses, tone changes, and real-time interaction with the speaker, without relying on written text in the training process. The company explained in an announcement published on October 23, 2026, that 'Disa' learns communication patterns directly from audio signals, including speech rhythm, response timing, tone variations, and accompanying expressions. This approach involves training the model from scratch using recordings of human conversations, without using pre-trained models, differing from voice systems that rely on converting speech to text before processing and generating responses. The company noted that converting speech to text may lead to the loss of some audio signals related to emotions, hesitation, sarcasm, and subtle tone changes, elements that 'Block' seeks to preserve through its new model. 'Block' showcased experimental recordings featuring the model.