Block's Dessa holds live spoken conversations after training only on human speech
Block's small research lab released a research preview of Dessa, a speech-to-speech model trained from scratch on human conversation audio, with no text and no pretrained model. It is full-duplex, meaning it listens and talks at the same time, so it can react while you are still speaking. Block says it hears emotion and sarcasm that a speech-to-text step throws away. Early access goes first to a small group of AI safety researchers.