Block's Dessa unveils startlingly realistic voice model trained on a single NVIDIA GB200 NVL72
TheTuringPost · x · 2026-10-24
A new voice model from Dessa, part of Block, is drawing attention for its remarkably natural-sounding output. The standout detail: it was reportedly trained on just a single NVIDIA GB200 NVL72, making the realism-to-compute ratio especially striking.
Related event: Block's Dessa: voice-only trained full-duplex speech model runs locally(21 posts)→
More from Multimodal
- MOSS-Image Ecosystem Roundup: Alpha-Channel LoRA, Hi-Res Adapter, A1111-Style Prompts — optimisticalish · 2026-10-24
- How those viral AI Reels are made: start from a real reel, extend style with Fable — jaivinwylde · 2026-10-24
- Midjourney draws a dino getting flipped off by a chicken, Nano Banana animates the sequel — michaelrabone · 2026-10-24
- Cosmic Watercolor Fusion: AI art blends fashion photography with fiery nebulae — LudovicCreator · 2026-10-24
- Kling 4.0 launches on Hailuo: reference-based scenes, 30s clips, 1080p — aziz4ai · 2026-10-24
- Interactive Demo Link Released for the GPT 6 Astra ISS WebGL Scene — ChrisGPT · 2026-10-24