Spoken Language Understanding (SLU) is a crucial task in spoken language processing. Despite the availability of numerous English datasets for research, there is a scarcity of resources for low-resource languages like Vietnamese. This paper introduces VN-SLU, the first dataset explicitly designed for Vietnamese SLU. VN-SLU includes 17,321 utterances from 240 Vietnamese speakers, obtained through novel crowd-sourcing methods. We propose a web tool for scenario generation and label validation, ensuring dataset quality and diversity. This tool prompts participants to confirm intents and slot values in smart home and virtual assistant dialogues, ensuring precise alignment. Experimental results highlight the challenging nature of the chosen test set sampling strategy in intent accuracy, SLU-F1, and utterance accuracy. Additionally, we explore the integration of pitch information into the Vietnamese SLU system. Results show improved performance compared to the baseline model.