Dessa is a new, authentic full-duplex speech model developed by Block that was trained entirely on human conversation without using any text or pretrained models. It is designed to naturally understand and express human emotions, nuances, and real-time timing through audio-to-audio interaction. Operating locally with fast inference, the model aims to improve human-AI communication and safety by truly hearing the emotional content of speech.
You learned to talk by listening. So did Dessa. A research preview of a full-duplex speech model.
