When investigating the ecological and behavioural patterns of wildlife through sound, bioacoustic studies using machine learning (ML), such as convolutional neural networks (CNNs), are key for analysing large acoustic datasets. For other biodiversity monitoring methods like ecosystem accounting, practitioners often do not have the required technical knowledge. This thesis presents a series of studies focused on the application of ML methods to bioacoustic data, and providing tools to assess ecological impacts. We begin by conducting a systematic literature review of Passive Acoustic Monitoring (PAM) to investigate how it’s used with ML methods and identify trends or gaps present. We find increases in dataset size over time, and spectrograms being the most popular representation of bioacoustic data for inference, but highlight the need for standardized evaluation methods and broader use of open-source to advance the field. Our second contribution is NEAL, an open-source Shiny R application, which enables granular annotation of audio data for use in training and evaluating species classification models. Its no-code interface and modular design empower use by non-programmers, improving annotation workflows and supporting machine learning model development in bioacoustics. Later, we use generative AI, specifically Stable Diffusion, to create synthetic spectrograms for use in training bird species classification models. We use a dataset annotated using NEAL to benchmark these models. We demonstrate that supplementing training data with synthetic samples enhances classification performance on the human-labelled test set. Our next contribution is ExActR, an open-source Shiny R application which enables environmental project managers to quantify land cover changes using geospatial datasets. It supports ecosystem extent accounting without requiring GIS expertise. ExActR facilitates accessible and reproducible ecological assessments, with potential to expand into a comprehensive tool for ecosystem accounting. Our final contribution investigates the use of a lightweight CNN to classify sparse vocalisations of the invasive small Indian mongoose in Okinawa using audio from camera trap videos. The classifier was then applied to a large acoustic dataset to gain insights into the distribution of the mongoose across Okinawa and aiding conservation efforts.