Block Unveils Dessa
The genuinely interesting part is local inference: a full-duplex speech-to-speech model running on a MacBook Pro puts Block in the same direction as Kyutai's Moshi — speech-only training and low-latency conversation without a cloud round-trip. But with no weights, no benchmarks and no size disclosed, Dessa is a demo, not a release, and every claim currently rests on Block's word.… more
It is also worth remembering that Block is a payments company, so this most likely targets its own customer-facing voice stack rather than the open model market. If weights ever ship, you can expect independent evaluations to show whether this is real progress or a well-produced showcase.
Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech
Block’s AI lab released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio. It runs locally on a MacBook Pro M4 Max, but Block has published no size, benchmarks or weights. The post Block Unveils Dessa, a Full-Duplex Speech Model Trained Only o…
Source:
Marktechpost
Discussion (0)