Key facts
- Dessa is a full-duplex voice model that runs locally on consumer GPUs.
- The model mimics human emotions like laughs, pauses, and cries.
- Dessa is trained entirely on human speech.
- Block is seeking safety researchers to test the model.
- The model can run on devices like a MacBook M4 Max or better.
Block, the company co-founded by Jack Dorsey, has unveiled Dessa, a new voice model designed to mimic human speech with emotional nuance. Unlike many current models that rely on text-to-speech pipelines, Dessa is a full-duplex model trained entirely on human speech and can run locally on consumer-grade hardware. Block claims the model can converse naturally, incorporating elements like laughter, pauses, and even crying, making interactions feel more authentic. The company is actively seeking safety researchers to evaluate Dessa for potential misuse, particularly in scenarios where voice agents might deceive users. Block, which already utilizes voice AI in its Square offerings, sees Dessa as a technology that could enhance various business and consumer experiences at a reduced cost.