Deep Dive into ByteDance Doubao's Duplex Model Seeduplex

ByteDance Doubao has unveiled a model called Seeduplex, which is a new duplex voice model. The review highlights its capabilities for parallel perception and interruption, as well as recognizing other voices. The model supports voice input, simplex and duplex modes; it can simultaneously recognize sounds, including knocking on a table. Seeduplex can mimic voice and style (e.g., a commanding tone with "forceful" endings), as well as imitate a gentle voice and add a hint of a smile. It can switch between male and female voices on request and change tone according to instructions. Additionally, the model supports mixed Chinese-English, recognizes most Chinese accents (e.g., some "Northern" speech), and can read handwritten texts with various fonts.