Jack Dorsey's Block Unveils Human-Like Voice AI Model Dessa

Financial technology corporation Block unveiled an experimental artificial intelligence model named Dessa, engineered from scratch using exclusively human conversational recordings rather than written text. Project lead Ragavan Thurairatnam shared demonstration recordings highlighting the system's human-like voice cadence, emotional inflections, and conversational pauses.

Members of the lab began building deep learning models in 2012 and have developed generative speech models for more than eight years, according to Unite.AI.

Dessa was trained from scratch on sounds alone, without any text or pretrained models, learning to converse purely by listening, according to Unite.AI.

Block stated that Dessa learned language from hundreds of times fewer words than large language models and trained on a single NVIDIA GB200 NVL72 system.

Block's AI research lab announced Dessa on October 23, 2026, a full-duplex speech-to-speech model trained on human conversation, according to Unite.AI.

The engineering team described Dessa as a genuine speech-to-speech architecture that derived its entire language comprehension directly from auditory interactions. Demonstrated audio samples featured informal discussions regarding cinema directors, natural responses expressing emotional warmth, and comparative tests against OpenAI's GPT-Live-1 voice pipeline.

Technical Architecture and On-Device Execution

The developers reported that Dessa can execute entirely locally on consumer hardware, functioning on an Apple MacBook Pro equipped with an M4 Max chip. Researchers explained that traditional voice assistants flatten verbal nuances into text tokens, whereas Dessa attempts to replicate how young children acquire speech purely through hearing. The engineering group argued that contemporary speech systems remain blind to human emotional states and vocal distress, positioning Dessa as a step toward addressing safety gaps in vocal interfaces.

Research Access and Corporate Strategy

Block Chief Executive Jack Dorsey stated that intelligence tools are transforming organizational efficiency by enabling smaller, highly leveraged engineering teams to achieve superior product outcomes.