Wake‑word detection is the gateway for virtual assistants, shaping the smoothness of user interaction. This work extends the classic direct keyword spotting with a contextual trigger detector. After the wake word fires, a reasoning module decides whether the following speech is a user command or unrelated talk, enabling efficient and context‑aware engagement. To train the system, a controllable multi‑speaker synthetic conversation generator was built, producing a 62.3‑hour corpus that contains direct invocations, contextual follow‑ups and non‑addressed chatter. Experiments across diverse synthetic scenarios confirm the approach’s effectiveness. The code, dataset and pretrained models are released to support reproducibility and further progress in intelligent assistant research. Review