Block Unveils Dessa, a Full-Duplex Speech Model
Block introduces Dessa, a full-duplex speech-to-speech model trained exclusively on human conversation audio, capable of running locally on a MacBook Pro M4 Max.
Stock photo for illustration only, not from the actual event
- Block has released Dessa, a full-duplex speech model trained solely on human conversation audio.
- The system is engineered to run locally on a high-end MacBook Pro M4 Max laptop.
- The company has not yet published model size, weight parameters, or benchmark results.
Block's AI lab has officially introduced Dessa, a full-duplex speech-to-speech model. The core distinguishing factor of this new model is its training methodology, which relies exclusively on human conversational audio data to capture natural dialogue dynamics.
A major highlight of Dessa is its deployment capability. Rather than requiring massive cloud infrastructure, the model is built to run locally on consumer-grade high-end hardware, specifically operating on a MacBook Pro M4 Max.
Stock photo for illustration only, not from the actual event
Running advanced speech models locally on edge devices like the M4 Max highlights a broader industry shift toward privacy-first artificial intelligence. By processing audio conversations directly on the user's hardware rather than transmitting sensitive voice data to remote servers, applications can achieve lower latency and enhanced user data protection.
Despite the notable announcement, Block has kept technical details under wraps during this initial release. The organization has not yet disclosed the exact model size, benchmark evaluations, or weight parameters, leaving developers waiting for further official documentation.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment