Bland AI releases Tofu-1, a full-duplex speech-to-speech model for high-stakes calls
CurieuxExplorer · x · 2026-11-17
Bland AI has launched Tofu-1, an end-to-end speech-to-speech model built for high-stakes, real-time phone calls in poor conditions. A limited research preview is now open to enterprises.\n\nKey features:\n- Full-duplex: listens while it speaks; callers can interrupt or correct it mid-sentence, without a traditional turn-taking loop\n- Direct audio processing: skips transcription-first pipelines, cutting latency and preserving vocal nuance\n- Noise resilience: handles moving cars, hospital rooms, weak signal, background TV, and multiple simultaneous speakers\n- Thinks while talking: keeps the call moving while retrieving information, never claiming a task is done early\n- Scene generation & speaker separation: tells speakers apart, separates voices from noise, and can generate ambient sound
More from Multimodal
- RMBG-3.0 launches as the #1 background removal model, beating remove.bg — bdsqlsz · 2026-11-17
- ByteDance's Seedance 2.5 now offers 4K video generation via API — aftahi_ai · 2026-11-17
- FLUX.1-schnell, Apache-2.0 open text-to-image model, trends again on HF — black-forest-labs · 2026-11-17
- Mirelo launches agentic sound-design tool that swaps music while preserving original SFX — HeyNayeem · 2026-11-17
- AI-Generated Film BY AND BY Releases Teaser Based on the Staircase-in-the-Woods Legend — Merzmensch · 2026-11-17
- Same Prompt Showdown: Seedance 2.5 vs MiniMax Kong Street Scene — Ok_Variety2532 · 2026-11-17