AI 导读
合在一起，这才是基础设施与演示的区别。懂你领域的识别能力、一套栈覆盖所有延迟目标，以及把上下文当作一等输入而非事后补充的设计。 来自网易有道的开源。 核心问题是：稳定、领域感知的流式识别是否会成为语音系统的常态。
正文
Together, that's what separates infrastructure from a demo. Recognition that knows your domain, one stack for every latency target, and a design that treats context as a first-class input rather than an afterthought.
Open source from NetEase Youdao.
The main question is whether stable, domain-aware streaming recognition becomes normal for voice systems.
来源：@rohanpaul_ai · x.com