Voice

Block’s Dessa asks what a voice model loses when speech becomes text

Block presents a speech-trained, full-duplex model and local Mac inference. The demos make a design proposition—not an independent reliability or empathy benchmark.