Bland releases the Tofu-1 speech model

Bland has released a full-duplex voice-to-voice model Tofu-1, used for calls in mobile cars, hospital rooms, and weak signal lines, directly processing audio input and output without converting to text, supporting listening while speaking, reasoning, and responding, with the first batch of audio skills including speaker tracking, noise separation, and paralinguistic recognition. Currently, only an enterprise research preview version API is open, targeting the banking, healthcare, insurance, and government industries, with application requiring monthly call volume (from 1,000 to 50,000+).

AI-written from the reports · Updated 7 hours ago