🤖Block Unveils Dessa: A Human-Sounding Voice Model
Dessa: The Most Realistic Voice AI Yet
TL;DR
Block's research lab released Dessa, a voice model that runs locally and sounds human with expressions like laughs and pauses. It's designed to be more natural and less annoying than current voice AI.
Block's research lab just dropped Dessa, a voice model that runs locally and sounds shockingly human. It can laugh, pause, and even cry, making it the most authentic full-duplex speech model ever built. Why should you care? If you're building voice interfaces for businesses or consumers, Dessa could be a game-changer. It's trained on human speech and runs on a single NVIDIA GPU, making it accessible to developers. The model's knowledge comes from conversational data, and it can run on consumer-grade computers like a MacBook M4 Max. But here's the catch: there's no public access yet, and Block is calling for safety researchers to test it.
Key Points
Dessa, Block's new voice model, runs locally on a single NVIDIA GPU and sounds human with expressions like laughs and pauses.
Dessa is trained fully on human speech, making it the most realistic full-duplex model ever built.
The model can run on consumer-grade computers like a MacBook M4 Max or better, making it accessible to developers.
Block is calling for safety researchers to test Dessa, as there's no public access to the model yet.
Square already uses voice AI to handle ordering, and Dessa could power similar business and consumer experiences at a lower cost.
Why It Matters
If you're building voice interfaces for businesses or consumers, Dessa could be a game-changer. It's trained on human speech and runs on a single NVIDIA GPU, making it accessible to developers. However, the lack of public access and the need for safety testing means it's not ready for prime time just yet.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,567 builders reading daily.