In superconducting magnets, the irreversible transition of a portion of the coils to the resistive state is called a quench. Having large stored energy, quenches can lead to damage of magnet components due to localized heating, high voltage, or large mechanical forces. Unfortunately, current quench protection systems can only detect the quench after it happens, giving magnet operators very short response time (a few ms). Furthermore, magnet quench's behavior is still not well-understood, which is why there has been significant uncertainty as to whether quench precursors exist, and in what kind of data they would appear. In this talk, I will discuss several machine learning techniques we are investigating for anomaly detection in superconducting magnets' acoustic data. Results from our experiments provide strong indications of quench precursors's existence in acoustic data. While the exact nature of the events are still not fully understood, we show that the system can utilize the events to forecast quenches seconds before they appear under magnet training conditions. Eventually, we would like to utilize our expertise in deploying real-time machine learning models to implement these techniques on hardware as real-time quench detection systems.