AIMPACT message, April 9: ByteDance's Seed team has launched the native full-duplex voice large model Seeduplex, which is now fully available on the DouBao app, marking a transition of voice interaction from "turn-based" to real-time natural conversation.
Seeduplex achieves synchronized "listen-and-speak" processing through joint acoustic and semantic modeling, significantly improving interference resistance in complex environments. Data shows that, compared to traditional half-duplex solutions, its misresponse and misinterruption rates are reduced by approximately 50%.
In terms of interactive experience, the model introduces dynamic stopping technology, reducing response latency by approximately 250 milliseconds and decreasing talk-over incidents by 40%, enabling more accurate differentiation between user pauses and conversation endings. Meanwhile, through speculative sampling and quantization optimization, the system maintains low latency and smooth performance even under high concurrency, increasing overall call satisfaction by approximately 8.34%.
This upgrade signifies that AI voice is evolving toward "real-time, multimodal, human-like interaction," and in the future, it is expected to integrate visual capabilities to advance intelligent assistants toward an integrated system of "listening, seeing, thinking, and speaking."(Source: ByteDance)