Block releases Dessa, a local full-duplex speech model, and seeks safety researchers to test it for deception risks
Block, Jack Dorsey's company, released Dessa from its research lab. It is a full-duplex speech-to-speech model trained entirely on human speech, and it produces human-like laughs, pauses and emotional expression. Block says it runs locally on consumer hardware such as a MacBook M4 Max or better. There is no public access, and Block has issued a call for safety researchers to test the model. TechCrunch suggests this is likely aimed at cases where a voice agent could trick the person on the other end. The article describes no vulnerability, exploitation or incident, so the relevance for defenders is the potential for more convincing AI-driven voice impersonation and social engineering.
Detalhes técnicos
- Dessa is full-duplex, so it can listen and talk at the same time.
- It does not use a text pipeline driving speech generation. Block says emotion is built into the model, though it is unclear whether tags steer it.
- Block says it was trained on a single NVIDIA GPU.
- Block's own blog post calls current voice AI "potentially dangerous" because it discards non-text information.
Linha do tempo
-
Block releases Dessa, a local full-duplex speech model, and seeks safety researchers to test it for deception risks
ler a análise desta fonte ocultar
Block, Jack Dorsey's company, released Dessa from its research lab. It is a full-duplex speech-to-speech model trained entirely on human speech, and it produces human-like laughs, pauses and emotional expression. Block says it runs locally on consumer hardware such as a MacBook M4 Max or better. There is no public access, and Block has issued a call for safety researchers to test the model. TechCrunch suggests this is likely aimed at cases where a voice agent could trick the person on the other end. The article describes no vulnerability, exploitation or incident, so the relevance for defenders is the potential for more convincing AI-driven voice impersonation and social engineering.
Technical Details
- Dessa is full-duplex, so it can listen and talk at the same time.
- It does not use a text pipeline driving speech generation. Block says emotion is built into the model, though it is unclear whether tags steer it.
- Block says it was trained on a single NVIDIA GPU.
- Block's own blog post calls current voice AI "potentially dangerous" because it discards non-text information.