Subglacial seismicity provides the opportunity to monitor inaccessible glacial beds at the epicentral location and time. Glaciers can be underlain by rock or till, which determines the mechanics of slip and, if unstable, characteristics of resulting seismicity. Utilizing a double direct shear apparatus, we found conditions for instability at freezing temperatures and high slip rates for both bed types, although with very different frictional evolution. During stick-slip stress-drops, we recorded acoustic emissions with piezoelectric transducers frozen into the ice. Supervised machine learning can classify recorded waveforms and spectra as coming from rock or till beds. The Random Forest Classifier is interpretable, with the prediction based on the initial oscillation peaks and high frequency energy. Till events are generally higher stress-drop, with more impulsive first arrivals compared to rock waveforms. These seismic signatures of mechanical slip processes and associated bed conditions can potentially greatly enhance interpretation of subglacial seismic data.