Dessa: the most authentic full-duplex speech model ever built
Summary
Dessa is Block's true speech-to-speech, full-duplex model trained from scratch on human speech with no text or pretrained models. It can listen and speak in real time with emotion and backchanneling, running locally on Mac hardware (MacBook Pro M4 Max or better). Block showcases live samples including conversations with GPT-Live-1 and invites safety researchers, emphasizing safety, privacy, and potential applications for natural voice interfaces.