Block Unveils Dessa Speech-to-Speech Model for Local MacBook Use

Block unveiled Dessa, a speech-to-speech AI model designed to listen and speak at once, with faster-than-real-time local operation on MacBook Pro M4 Max systems and higher-spec devices.

Block’s AI research lab trained the model from scratch on human conversation audio, without text or pretrained models. The training used one NVIDIA GB200 NVL72 system.

Dessa uses full-duplex speech, so it can take in audio while producing a spoken response. Block has shared demonstration recordings and granted early access to a small group of AI safety researchers. The model remains a research preview and is not broadly available.

Project lead Ragavan Thurairatnam described Dessa as a GPT-3-like moment for speech AI.