Water loss in water distribution networks (WDNs) is a multi-billion-dollar global issue. WDNs often develop leaks and breaks over time due to factors such as aging infrastructure, pressure transients, and operational changes. Leaks in WDNs are a critical area of concern for utilities as they contribute to water loss, increase the risk of waterborne pathogens, and pipe breakage. Additionally, they are challenging to find and repair given the subterranean nature of most WDNs. To reduce the risks and impacts of leakage, it is necessary for utilities to constantly monitor, detect, localize, and repair leaks in a timely manner. Due to the scale and complexity of municipal WDNs coupled with resource constraints at the operational level, effective leak detection approaches need to be robust, efficient, and easily deployable across networks. Currently, most leak detection methods have common deficiencies with respect to both sensing and analytical approaches that significantly limit their effectiveness in practice: vibration-based methods are not conducive in PVC networks; and the majority of analytical approaches are contingent on the availability of accurate, auxiliary information—such as network layout and pipe specifications—for detection and localization. To address these gaps, this paper focuses on tandem development across two aspects: improving the empirical understanding around hydrophones for leak detection; and further development of autonomous (i.e., self-sufficient) algorithms for real-time leak detection using machine learning and signal processing.