HomeNews

Bland Launches Tofu-1 Speech Model for Noisy Phone Calls

Kamran
Kamran
•
Nov 16, 2026 12:33 PM
•
0
 min read
Select Emergent as your Preferred news source
Bland Launches Tofu-1 Speech Model for Noisy Phone Calls

💡 TL;DR

  • Bland AI launched Tofu-1 on November 15, 2026, a full-duplex speech model that processes phone audio directly without text transcription.
  • The model isolates caller voices from background noise like crying children, traffic, TVs, and intercoms during live calls.
  • Enterprise preview is live for banking, healthcare, insurance, and government customers with general availability planned for January 2027.

Bland AI officially launched Tofu-1 on November 15, 2026, introducing a full-duplex speech-to-speech model designed for high-stakes phone conversations in noisy environments. The model processes audio directly without converting speech to text, targeting enterprise use cases in banking, healthcare, insurance, and government where background interference is common.

What Tofu-1 Does Differently

Tofu-1 skips the conventional pipeline of transcribing caller audio to text, processing it through a language model, and synthesizing a voice response. Instead, the model operates as a native speech-to-speech system that listens, reasons, and speaks simultaneously. According to Bland's product page, the model maintains full-duplex capability: it can adjust what it is saying based on what it hears in real time.

The company positions Tofu-1 for scenarios where audio quality is already compromised. Bland states the model is built for calls from moving cars and hospital rooms, over weak signal, with several people talking at once. The system reportedly isolates a caller's voice from competing sounds including televisions, crying children, traffic, intercoms, and keypad tones. When background noise makes a caller impossible to understand, the model is designed to acknowledge the issue and ask the caller to repeat themselves or move to a quieter location.

Release Date and Availability

Tofu-1 officially launched on November 15, 2026, in a gated enterprise research preview. Access is currently limited to a select group of organizations in banking, healthcare, insurance, and government sectors. Bland AI has announced general availability for January 2027, though no specific date within that month has been disclosed.

Technical Details and Benchmarks

Bland has not published benchmarks, model size specifications, or pricing information for Tofu-1. The company describes the release as the first in a planned family of speech-to-speech models, suggesting additional versions will follow. The absence of public benchmarks means comparative performance against other full-duplex models remains unverified.

The launch arrives two days after inclusionAI released Duo v1, another native full-duplex model, indicating growing momentum in the speech-to-speech model category. Both releases suggest enterprises are seeking alternatives to traditional text-mediated voice AI pipelines for real-time conversational applications.

What This Means

Tofu-1 represents Bland AI's entry into native audio processing for enterprise telephony, targeting a segment where traditional speech recognition struggles with environmental noise. The gated preview structure and enterprise focus signal a cautious rollout aimed at refining the model in controlled, high-value use cases before broader release. Whether Tofu-1 delivers on its noise isolation claims will depend on field performance data that Bland has yet to share publicly. For now, the model offers enterprises in regulated industries a new option for handling difficult audio conditions without relying on text intermediaries.

About the writer

Kamran Alam is an Engineer at Emergent, where he builds the core platform powering AI-driven software creation. He previously co-founded Jeevam Health (YC S20) as CTO and holds a B.Tech. from IIT Roorkee.

Start Building
on Emergent today
Try Emergent