marktechpost.com web signal

Bland AI Debuts Tofu-1 Full-Duplex Voice Model in Preview

Voice AI Generative AI ai-business

TL;DR

  • Bland AI opened a research preview of Tofu-1 on November 15, a native full-duplex speech-to-speech model for high-stakes phone calls.
  • Access is limited to enterprises in banking, healthcare, insurance and government; general availability is slated for January and no benchmarks have been published.
  • Tofu-1 processes audio directly instead of transcribing it first, pitched to separate a caller from a TV, a crying child, traffic, an intercom or keypad tones.

Bland AI opened a research preview of Tofu-1 on November 15, a speech-to-speech model it pitches for phone calls where the audio is already bad.

On its product page, the company describes Tofu-1 as "full duplex: it listens and speaks at the same time, and what it hears can change what it is saying." Per MarkTechPost, it skips the usual step of transcribing a caller to text, running a language model over the transcript, and then voicing a reply. Tofu-1 processes the audio directly.

The target conditions are specific. Bland says the model is built for "calls from moving cars and hospital rooms, over weak signal, with several people talking at once," and that it separates a caller's voice from a TV, a crying child, traffic, an intercom or keypad tones. It also promises the model should "say so and ask them to repeat themselves or move somewhere quieter" when the background makes a caller impossible to understand.

The preview is gated. Access is open to a limited number of enterprises in banking, healthcare, insurance and government, with general availability slated for January. Bland has not published benchmarks, model size, or pricing, and Tofu-1 is the "first release in Bland's family of speech-to-speech models."

The release lands two days after inclusionAI's Duo v1, another native full-duplex launch our voice-AI tracker picked up earlier in the week.