Block, led by Jack Dorsey, has introduced a new full-duplex speech-to-speech model called Dessa. This model is entirely trained on human voice data, supports local deployment, and delivers realistic pronunciation with the ability to simulate a wide range of emotions. Its knowledge base is directly derived from conversational data, eliminating the need to rely on large language models for knowledge retrieval. This effectively addresses the issue of existing voice AI systems often losing emotional information and nuances in tone during interactions. Dessa can operate efficiently on consumer-grade devices or a single NVIDIA GPU rack. Currently, Block has not fully opened the model to the public. Instead, it has initially invited security researchers to participate in testing to evaluate its performance across various scenarios. The launch of Dessa is expected to provide businesses and end-users with a more cost-effective, immersive, and natural intelligent interaction experience, propelling voice AI towards a new phase of native voice interaction.