Models New
Bland opens a research preview of Tofu-1, a full-duplex speech-to-speech model for phone calls
Published:
San Francisco voice-AI company Bland has introduced Tofu-1, which it describes as a full duplex speech-to-speech model built for calls in poor conditions: from moving cars and hospital rooms, over weak signal, or with several people talking at once. Bland announced it on 13 November, and MarkTechPost reported it on Sunday.
Bland says Tofu-1 listens while it speaks, so callers can interrupt or correct it mid-sentence, keeps the call moving while it works on a task, tells speakers apart and can generate several speakers and the sound of the room. Unlike cascaded voice systems that turn speech into text and back, it works on the audio itself, the company says.
Access is a research preview for a limited number of enterprises. Bland has not published the model's internal architecture, and its parameter count and turn-taking latency are not disclosed, MarkTechPost notes.
Caveat. All capability claims are Bland's own; no independent evaluation or benchmark result was available, and the model is not generally available.
Sources