Brief
Block unveils speech‑first voice model Dessa for on‑device AI
Dessa processes spoken input directly, enabling real‑time, interruption‑aware dialogue on high‑end laptops.
By Felo News Desk · Published
Block introduced Dessa, a speech‑first voice model that processes audio without converting it to text first, allowing users to talk over the system and maintain a natural back‑and‑forth, Trendhunter reported on Oct. 26.
What happened
Dessa was built from human conversation data and uses a full‑duplex architecture so it can listen and respond simultaneously. The model runs locally on MacBook Pro machines equipped with M4 Max or higher chips, eliminating the need for cloud infrastructure.
What the reports add
Trendhunter notes that early access is limited to a small group of AI safety researchers, and that the on‑device deployment creates a test bed for evaluating voice‑AI safety without external servers.
What was said
Block’s description emphasizes that the duplex speech approach “shifts the category beyond text‑based call and response, making interruption handling and vocal nuance part of what teams evaluate in voice AI.”
How it came about
Block’s launch follows a series of recent voice‑AI developments, including full‑duplex models highlighted in earlier Felo coverage of voice‑AI breakthroughs.
Key facts
- Block released the speech‑first AI model named Dessa. (trendhunter.com)
- Dessa runs locally on MacBook Pro M4 Max or higher hardware. (trendhunter.com)
- The model uses full‑duplex architecture to listen and speak simultaneously. (trendhunter.com)
- Early access is limited to a handful of AI safety researchers. (trendhunter.com)
Sources
- [1] trendhunter.com — originally reported as “Speech-First Voice Models”
