Block's Dessa: Full-Duplex Voice Model With Laughs and Pauses Runs Locally on a MacBook
SumitGup · x · 2026-10-25
Block, Jack Dorsey's company, released Dessa, a full-duplex speech model from its research lab that sounds "eerily human"—complete with laughs, pauses, cries, and natural cadence.
- Full-duplex: talks and listens simultaneously, comparable to OpenAI's GPT-Live;
- Trained purely on speech: knowledge comes from conversational data rather than a paired LLM; Block argues existing voice models sound unnatural because text sits in the pipeline;
- Low compute bar: trained on a single NVIDIA GPU rack, and it can run locally on a MacBook M4 Max.
Related event: Block unveils Dessa, a speech-only full-duplex voice model(30 posts)→
More from Multimodal
- No Need to Hate AI Filmmakers, It's Just a New Genre: BBC Radical Interview Sparks Debate — DavidmComfort · 2026-10-25
- Neo-Noir AI Film 'Silence Speaks' Tributes The Silence of the Lambs via Sora and Suno — Potential-Program964 · 2026-10-25
- AI-Generated Series 'Bambi the Destroyer' Releases Episode 18 — Severe-Tank-5252 · 2026-10-25
- Aggregated 18 ComfyUI GPU-unlock benchmarks show BF16 gains dwarf INT8 — Acceptable_Put9983 · 2026-10-25
- Fully AI-generated Green Day cover: one continuous take from Yue2, Krea2 and FL2V — Nitely_Productions · 2026-10-25
- BRIA's Fibo-2.0-editing Image Editing Model Trends on Hugging Face — briaai · 2026-10-25