Monitoring freshwater species using conventional methods is often labour-intensive, invasive, and expensive. As a result, ecologists are now supplementing conventional methods with new cutting-edge technologies, such as passive acoustic monitoring, which involves recording all of the sounds in an environment to address ecological questions. Passive acoustic monitoring is rapidly becoming a gold standard for affordable non-invasive ecosystem monitoring at large spatial and temporal scales as many inexpensive acoustic sensors can be deployed for months at a time in many locations simultaneously. Passive acoustic monitoring can also provide information regarding behavioural and physiological status. However, passive acoustic monitoring generates vast amounts of data and requires innovative computational solutions to derive meaningful ecological conclusions. Machine learning models, such as convolutional neural networks and transformers are state-of-the-art techniques used in the classification and detection of species-specific sounds. Convolutional neural networks and transformers have been used to successfully detect bird and frog species in rainforests, and whale echolocation clicks in the ocean. There is huge untapped potential in passive acoustic monitoring to harness recent advances in machine learning to provide inexpensive and effective freshwater ecological assessment at large spatial and temporal scales. Despite this, many of the sounds detected in freshwaters are largely unknown and cannot yet be directly linked to the species that produced them. A new initiative, The Freshwater Sounds Archive, is beginning to catalogue species-specific sounds from around the world, which will assist in the development of species classifiers using machine learning. In this talk, we push past the bulrushes and dive below the lily pads to explore this new sonic frontier and see how recent advances in machine learning might drive research and help decode the mysteries of the freshwater soundscape.