Analysis of earthquake spectra can aid in understanding source characteristics like stress drop and rupture complexity. There is growing interest in probing the similarities and differences of fault rupture for natural and human-induced seismic events. Here, we analyze waveform data from a shallow, buried geophone array that recorded seismicity during a hydraulic fracturing operation near Fox Creek, Alberta. Starting from a quality-controlled catalog of 4000 events between magnitude 0 and 3.2, we estimate source-spectral corner frequencies using methods that account for the band-limited nature of the sensor response. The stress-drop values are found to be approximately self-similar, but with a slight magnitude dependence in which larger events have higher stress drop (∼10 MPa). Careful analysis of the relative corner frequencies shows that individual fault and fracture segments experienced systematic variations in relative corner frequency over time, indicating a possible change in the stress state. Clustering analysis of source spectra based on the relative proportion of high- and low-frequency content relative to the Brune model further shows that event complexity evolves over time. In addition, the faults produce earthquakes with systematically larger stress-drop values than the fractures. Combined, these results indicate that the features activated by hydraulic fracturing experience observable changes in source behavior over time and exhibit different properties depending on the orientation, scale, and fabric of the structural feature on which they occur.