Existing audio-based asthma monitoring solutions rely on feature engineering designs paired with contact-based auscultation which are brittle in practice and do not scale beyond point of care setups. Data-driven methods utilizing contactless microphones have the potential to address such limitations. These solutions are under-explored in healthcare due to high cost of data curation requiring physicians-in-the-loop. Here, we propose an active learning (AL) system to facilitate audio data collection and annotation. It detects lung sound abnormalities in asthma. AL reduces the annotation cost while increasing the model performance under a constrained annotation budget. It automatically extracts interesting audio segments from the continuous recordings, and efficiently annotates and trains anomaly detector model. The experimental results confirm the effectiveness of the proposed system as an enabler for larger scale data curation on a newly collected audio corpus for pediatric asthma.