Bland Launches Tofu-1, a Full-Duplex Speech-to-Speech Model for Calls

What happened

Bland introduced Tofu-1, a full-duplex speech-to-speech model designed to handle noisy phone calls in real time. It targets high-stakes conversations where both parties talk simultaneously, enabling seamless two-way voice translation or enhancement. The model is currently in enterprise research preview, so details on performance benchmarks, model size, and pricing have not been disclosed.

Why it matters

Phone calls remain a critical communication channel for many businesses and organizations, especially in sectors like customer support, finance, and emergency services where clarity and timing are crucial. Traditional speech-to-speech systems typically process one speaker at a time, creating delays and making natural conversations challenging in noisy environments. Tofu-1’s full-duplex approach aims to remove this bottleneck by processing both sides of a conversation simultaneously, reducing lag and potentially improving call quality and comprehension under harsh acoustic conditions. This can lower friction, save time, and reduce misunderstandings in high-pressure call scenarios.

What to watch next

Operators should track Bland’s next updates to understand Tofu-1’s practical readiness, including published benchmarks and integration options. Pricing and enterprise licensing terms will shape adoption among contact centers, translation services, and business communication platforms. Watching how Tofu-1 performs in real-world noisy environments versus existing solutions will reveal if full-duplex models can shift call technology standards or remain niche tools. Also, competitors could accelerate development in duplex speech models, applying pressure on providers relying on sequential or half-duplex processing.

AI Quick Briefs Editorial Desk