SiC/SiC composites are refractory, damage tolerant, high-strength materials that are ideal for use in turbine engines. Although these properties allow engineers to design aircraft with better fuel efficiency, the safety-critical nature of their application space necessitates a detailed understanding of how damage initiates and progresses. Recent advances in machine learning (ML) and related statistical tools have created novel characterization pathways for understanding damage accumulation, however their widespread adoption is limited due to a lack of interpretability. This work lays the foundation for trustworthy ML for interpreting acoustic emissions (AE) that are produced when SiC/SiC composites sustain damage. A modified signal representation scheme was combined with unsupervised clustering to form the spectral framework. This represents the first AE-ML approach that can distinguish between fiber break and matrix crack signals in SiC/SiC minicomposites. The spectral framework was then benchmarked against 4 state-of-the-art AE-ML frameworks and shown to achieve superior performance. Community guidelines for standardized benchmarking were proposed to promote greater transparency. Finally, an in situ x-ray computed tomography experiment demonstrated that matrix crack signals overlap with early fiber breaks, and prevents the use of unsupervised clustering for realistic composite geometries. An autoencoder framework was created to overcome this limitation and used to demonstrate that the frequency distribution of fiber break signals is compact. These findings establish a pathway for real-time health monitoring.