🤖Block's Dessa Talks and Listens at Once on One GPU
TL;DR
Block's research lab unveiled Dessa, a full-duplex speech-to-speech model that talks and listens at the same time. It was trained only on human speech on a single Nvidia GPU and runs locally on a MacBook M4 Max.
Block's research lab unveiled Dessa, a full-duplex speech-to-speech model that talks and listens at the same time. It was trained only on human speech on a single Nvidia GPU and runs locally on a MacBook M4 Max. The catch: it isn't public yet.
Key Points
Dessa is full-duplex, so it can interrupt and be interrupted mid-sentence
Block says training used a single Nvidia GPU and human speech only, with no paired LLM
Runs locally on consumer hardware such as a MacBook M4 Max or better
Not publicly released; safety researchers are testing it for deceptive voice-agent use
Competes with OpenAI's GPT-Live, ElevenLabs v4 and Fish Audio Drama 3
Why It Matters
A voice model that runs on a laptop and needs no LLM backbone cuts latency and cloud cost for small-business voice ordering. Watch whether Block ships weights or keeps it as a Square feature.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,567 builders reading daily.