Block’s Dessa Speech Model Runs Locally on an M4 Max Mac

Speech in, speech out, with no text in between, and small enough for a laptop.

@kimmonismus says Jack Dorsey released a speech model that sounds best in class, and that “even more interestingly” it was trained on only one NVIDIA GB200 NVL72. “Oh, and it runs locally on your Mac,” he adds.

Block’s page describes Dessa as a true speech-to-speech model. It was trained from scratch on human conversation only: no text and no pretrained models. It is full-duplex, so it listens and talks at the same time, and it learned when to cut in and when to say “yeah” to show it is following. Inference runs faster than real time, Block says, on a MacBook Pro M4 Max or better.

The lab behind it is a small research team at Block, founded by former members of Dessa, the Toronto deep learning startup, and reporting to Dorsey. @jack announced the model on Thursday. “Seriously mindblowing,” @kimmonismus wrote on Saturday. “Feels like magic.”