Block Launches Dessa, a Speech-Only AI Model Trained Without Text Like a Learning Infant

Block, the payments and financial services company founded by Jack Dorsey, unveiled Dessa on October 23, 2026, a full-duplex speech-to-speech AI model trained exclusively on human conversation audio without any text or pretrained models. The project represents a departure from conventional voice AI architectures, which typically convert speech to text, process text, then generate new speech. Instead, Dessa was developed from the ground up on audio alone—a training approach inspired by how human infants acquire language before learning to read.