AI Gets Better Ears: New Model Talks *and* Listens Without Typing

Bland, a San Francisco voice AI company, has launched Tofu-1, a new model that handles phone calls in a fundamentally different way. Instead of the common 'cascade' method—converting speech to text, processing the text, and then converting the reply back to speech—Tofu-1 is a 'full-duplex speech-to-speech' model. It listens directly to the audio stream, processing tone, background noise, and hesitation to understand the conversation without an intermediate text transcript.

This approach could make voice AI agents more natural in challenging environments like moving cars or noisy hospital rooms. The model is designed to handle interruptions, separate voices from background noise, and keep the conversation flowing while it thinks. For developers and product teams, this signals a shift toward more integrated, audio-native AI assistants that could feel more human-like on calls, moving beyond the robotic, turn-taking systems of the past.