Block releases Dessa, a full-duplex speech model trained exclusively on human audio

·2026.10.24 04:09

Key point

The model operates as a true speech-to-speech system with inference speeds faster than real-time and runs locally on MacBook Pro M4 Max or higher.

Block's AI research team released Dessa, a full-duplex speech model trained solely on human conversation audio without using text or pre-trained models.
The model processes audio directly to understand meaning, enabling natural expression of emotions, laughter, and tone shifts while maintaining inference speeds faster than real-time.
Training required hundreds of times less data volume than typical LLMs and was completed on a single NVIDIA GB200 NVL72 hardware unit.
Local deployment is supported on MacBook Pro M4 Max or higher, and early access is being provided to a small group of AI safety researchers to identify risks specific to audio-only models.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.