Acoustic emission (AE) is a key method in the field of fatigue, both as a high-performance NDT tool for structural monitoring and as a fine-tuned approach to physical investigation. To tackle the major issues of source discrimination and sensitivity to ambient noise, signal processing techniques need to be developed. The present contribution shows how AE signals associated with fatigue can be automatically detected and grouped using machine-learning methods in an unsupervised manner. Compared with conventional analysis, which involves extracting descriptors from individual AE signals, these signal processing detection and clustering approaches are based directly on the comparison of complete waveforms – individual signals – or on continuous, high-frequency sampled acoustic recordings. The study shows that data mining techniques are capable of extracting and characterizing the acoustic signatures of fatigue – such as crack propagation or crack surface impacts – from noise and signals coming from other seemingly irrelevant sources. Based on the DBSCAN algorithm combined with cross-correlation and hierarchical clustering, it is possible to perform automatic, unsupervised detection and classification of AE fatigue multiplets in service, irrespective of material, sensor or loading. A second approach, based on a deep scattering network working on continuous acoustic recordings, demonstrates that the signal background contains relevant information that can be extracted – even at very low signal-to-noise ratios -, grouped into source mechanisms and investigated to gain an in-depth understanding of fatigue crack activity (closure, crack tip plasticity, etc.)