Block’s Dessa is a full-duplex speech-to-speech model trained only on human conversation audio—no text pretraining. It supports interruptions, backchannels, and expressive prosody; inference is faster than real time and runs locally on MacBook Pro M4 Max or better. Training used one NVIDIA GB200 NVL72 system. Public access is waitlist-only with early access for AI safety researchers; no API, weights, or benchmarks published yet.