The viability and competitiveness of wind energy industry require novel maintenance strategies and condition monitoring systems. This work presents a non-destructive testing system formed by acoustic sensors embedded in unmanned aerial vehicles to acquire acoustic signals from rotatory components of the nacelle. The occurrence of several noises increases the complexity of the analysis, requiring different filtering and data processing techniques based on wavelet transform and Butterworth filters. This approach is tested with a real case study with acoustic data acquired by an aerial acquisition system from an operating offshore wind turbine. The absence of faults is solved including recordings from laboratory associated to faults that simulate the main performance of rotatory faults in the nacelle. The results after the processing and filtering process provide clearly differentiated scenarios that demonstrate the viability and reliability of pattern identification using this technology, providing quantitative data about the real state of wind turbines.