Block previews Dessa, a voice AI model trained without text

Jack Dorsey's Block is previewing Dessa, an experimental full-duplex speech-to-speech model trained only on recordings of human conversations, with no text or pretrained models. It was trained on a single NVIDIA GB200 NVL72 rack with 72 GPUs and can run locally on a MacBook Pro with an M4 Max or better.

• Trained from scratch on human conversation audio only, no text or pretrained models
• Full-duplex: processes audio directly without speech-to-text conversion
• Trained on one NVIDIA GB200 NVL72 system with 72 GPUs
• Runs locally on a MacBook Pro with M4 Max or better