Since the discovery of fatigue phenomena, scientific research has constantly sought to understand and anticipate the failure of materials due to fatigue to mitigate unforeseen accidents and malfunctions in various technical fields. Numerous studies using acoustic emission (AE) -a key method in non-destructive testing -have shown a correlation between acoustic activity and fatigue damage. However, these measurements suffer from the non-specific nature of AE signals, which may be due to various physical sources. To investigate further the mechanisms of AE emission associated with fatigue, we study the groups of acoustic signals generated by fatigue cracking in metals. These so-called acoustic multiplets are characterized by highly correlated waveforms, are repeatedly triggered over many successive loading cycles at nearby stress levels and originate from a single location. These acoustic signatures produced during the propagation of fatigue cracks in alloys are automatically detected by a dedicated algorithm, grouped into multiplets and analyzed to understand the physical mechanisms from which they originate. By synchronizing their detection with digital image correlation measurements of fracture mechanics quantities, the investigation of this acoustic emission phenomenon shows that two mechanisms are at the origin of the multiplets: repeated local friction over fracture surfaces, and incremental crack propagation in the Paris regime, probably due to the reactivation of crack tip plasticity at each cycle. These two multiplet types serve as acoustic signatures, distinctly indicating the existence and propagation of a fatigue crack.