Very sophisticated machine learning tools are being developed for detecting P and S-waves in tectonic earthquakes, with excellent results, especially when approached from a recurrent perspective. However, their application to volcanic seismicity presents challenges due to the low magnitude, variability, and complexity of waveforms, caused by heterogeneous and anisotropic geological structures like magma chambers, rock types, and fractured zones. The proximity of sources to sensors often results in nearly simultaneous arrivals of P and S-waves. Additionally, volcanic areas are associated with high levels of seismic noise from non-volcanic sources. The specific characteristics of each volcano further necessitate adapting solutions to their unique dynamic behavior. Given these challenges, investigating signal preprocessing techniques that can improve P and S-wave detection in volcanic environments is essential. In this work, we studied seismic signals from the Nevados del Chill & aacute;n volcanic complex to evaluate whether simple yet robust information could be provided to an LSTM model for effective P and S-wave detection. Our approach achieved 94% detection rate for P-waves and 91% for S-waves within a 0.5-second error margin, for 998 P and S-waves from the test set, improving detection accuracy and noise resilience over traditional methods.