工具VIDEO
有道 Confucius R2T2 实时语音模型开源
有道开源实时语音识别模型 Confucius R2T2，核心机制是决定「何时不预测」，以等待换准确率，现已开放源码。
Alex Prompter
@alex_prompter2026.09.21收录原推发布于 2026.09.20
Alex Prompter @alex_prompter原推发布于 2026.09.20
Most latency work in AI is about predicting faster. Youdao Confucius R2T2 does the opposite thing. It has a mechanism for deciding when NOT to predict. Now open source. https://t.co/kaa1Wkdj10
x.com/alex_prompter/status/2101698934389477471采集于 2026.09.21 10:24