The localization of seismic events is a relevant step for volcano monitoring as it provides information of physical parameters that support the forecasting of volcanic activity. Automating this step is necessary to process the enormous amount of signals received by the observatories, improving relevant and opportune information retrieval at the early stages of monitoring. Localisation of seismic events is based on identifying the arrival time of the P and S waves in the raw signals, in a procedure called P and S picking. This is mainly performed manually or semi-automatically, usually applying the STA/LTA algorithm. The literature shows that the automation research on this process has been focused mainly on tectonic seismicity since the P and S waves are more difficult to detect in volcanic seismicity. This work aims to automate this process, proposing a novel and simple technique for volcano seismic events, using spectro-temporal energy features extracted from the spectrograms. The energy accumulated by an event over time, its second derivative, as well as a modification of the STA/LTA method were applied to detect the P and S waves for volcano tectonic (VT) events and only the P wave for long period (LP) events. The proposed method was adjusted and tested with the multi-station events recorded at the active Nevados de Chillan Volcanic Complex (Chile) and the results were compared with the reference manual picking provided by the observatory and with the traditional STA/LTA approach. The spectro-temporal approach presented lower errors and standard deviations than STA/LTA, in particular for the P detection for LP and the S detection, which are more difficult to detect. The median time error for the P wave detection was 0.0430 ± 0.21s for VT events and 0.0870 ± 0.35s for LP. For the S wave detection the error was 0.05 ± 0.49s. 80% of the P wave detection absolute errors were below 0.13[s] for VT and below 0.48[s] for LP. In turn, for the S wave detection, 80% of the errors were below 0.47[s]. In addition to the time errors, the distance errors were also evaluated, thus the localization was calculated for each picking method. The results presented a longitudinal error of 0.3° for both methods and a latitude error of 0.01° for the spectro-temporal method and 0.03° for STA/LTA method, once again demonstrating the good performance of the proposal. These results are promising despite the complexity of the P and S wave detection in volcanic events and the simplicity of the method compared to the methods proposed in the literature.