This work shows a systematic review of literature of machine learning and audio processing for applications in preventive maintenance. It intends to prove how audio signal processing combined with Machine Leanring techniques can produce a powerfool tool to detect anomalies and malfunctions in electromechanical devices. The document describes a review of art state and the algorithms that can be used for preventive maintenance applications. When reviewed, the literature proves that Machine Learning and Deep Learning approaches provide high accuracy results as tools for PdM, and that Support Vector Machines, k-Nearest Neighbors and Convolutional Neural Networks are the most used approaches as they achieve the highest evaluation metrics, and prove sound, vibration and current to be the most popular signals to train ML-PdM oriented models.