A computer-implemented method for determining a belt tension of a drive belt of a machine, based on a sound recording, including the steps of initiating a recording of sound, recording the sound during a primary time interval, discretising the sound into time samples, thereby obtaining a data set, filtering the data set, determining an average excitation spectrum, compensating for the average excitation spectrum, determining a confidence coefficient based on a highest and second highest extremum, and when the confidence coefficient exceeds a predefined value, determining the belt tension based on a frequency of the highest extremum.