mt logoMyToken
ETH Gas
EN

字节跳动发布音视频全双工大模型SeedRealtime

Favoritecollect
Shareshare

PANews 8月5日消息,据财联社报道,字节跳动正式推出原生音视频全双工大模型SeedRealtime。SeedRealtime用统一架构原生融合音频、视频与文本,能在连续的多模态信息流上实时交互,带来“边看、边听、边说”的全新体验。端到端人工评测结果显示,相比级联模型,SeedRealtime的音视频对话节奏问题减少了一半——模型对说话时机的把握更加自然,“话未说完被抢断、话音已落却回应迟缓、或是被背景杂音与闲聊误触发”等卡壳现象显著减少;同时,单次对话能够完整、顺畅交流的概率也有明显提升。目前,SeedRealtime已在豆包App全量上线。

Disclaimer: This article is copyrighted by the original author and does not represent MyToken’s views and positions. If you have any questions regarding content or copyright, please contact us.(www.mytokencap.com)contact
More exciting content is available on
X(https://x.com/MyTokencap)
or join the community to learn more:MyToken-English Telegram Group
https://t.me/mytokenGroup